def get_user_feature(feature_type,behavior,num_feature=800):
X_train = get_features(feature_type,behavior)
index = X_train.index
# 对X进行降维
Y = pd.read_csv('data/train_Y_%d.csv'%behavior, index_col='user_id')['type']
print 'start selectKbest...'
# select = SelectKBest(chi2,k=min(num_feature,X_train.shape[1]))
percent = 0
if feature_type == 'cat_id':
percent = 60
elif feature_type == 'brand_id':
percent = 15
elif feature_type == 'seller_id':
percent = 20
select = SelectPercentile(f_classif, percentile=percent)
select.fit(X_train,Y)
X_train = select.transform(X_train)
print 'end select...'
print 'write %s features to train file' % feature_type
train_feature_file_name = 'data/train_feature_%s_%d.csv' % (feature_type,behavior)
DataFrame(X_train,index=index).to_csv(train_feature_file_name)
# 用同样的列降维对应的测试集数据
X_test = get_features(feature_type,behavior,is_train=False)
index = X_test.index
X_test = select.transform(X_test)
# 写入文件
print 'write %s features to test file' % feature_type
test_feature_file_name = 'data/test_feature_%s_%d.csv' % (feature_type,behavior)
DataFrame(X_test,index=index).to_csv(test_feature_file_name)
print 'end....'
def preprocess(words_file = "word_data.pkl", authors_file="email_authors.pkl"):
### the words (features) and authors (labels), already largely preprocessed
### this preprocessing will be repeated in the text learning mini-project
authors_file_handler = open(authors_file, "r")
authors = pickle.load(authors_file_handler)
authors_file_handler.close()
words_file_handler = open(words_file, "r")
word_data = cPickle.load(words_file_handler)
words_file_handler.close()
### test_size is the percentage of events assigned to the test set
### (remainder go into training)
features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(word_data, authors, test_size=0.1, random_state=42)
### text vectorization--go from strings to lists of numbers
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5,
stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
### feature selection, because text is super high dimensional and
### can be really computationally chewy as a result
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
### info on the data
print "no. of Enrique training emails:", sum(labels_train)
print "no. of Juan training emails:", len(labels_train)-sum(labels_train)
return features_train_transformed, features_test_transformed, labels_train, labels_test
def preprocess(words_file = "../tools/word_data.pkl", authors_file="../tools/email_authors.pkl"):
"""
this function takes a pre-made list of email texts (by default word_data.pkl)
and the corresponding authors (by default email_authors.pkl) and performs
a number of preprocessing steps:
-- splits into training/testing sets (10% testing)
-- vectorizes into tfidf matrix
-- selects/keeps most helpful features
after this, the feaures and labels are put into numpy arrays, which play nice with sklearn functions
4 objects are returned:
-- training/testing features
-- training/testing labels
"""
### the words (features) and authors (labels), already largely preprocessed
### this preprocessing will be repeated in the text learning mini-project
authors_file_handler = open(authors_file, "r")
authors = pickle.load(authors_file_handler)
authors_file_handler.close()
words_file_handler = open(words_file, "r")
word_data = cPickle.load(words_file_handler)
words_file_handler.close()
### test_size is the percentage of events assigned to the test set
### (remainder go into training)
features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(word_data, authors, test_size=0.1, random_state=42)
### text vectorization--go from strings to lists of numbers
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5,
stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
### feature selection, because text is super high dimensional and
### can be really computationally chewy as a result
# selector = SelectPercentile(f_classif, percentile=10)
## <Temporary hack for Lesson 3>
selector = SelectPercentile(f_classif, percentile=1)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
### info on the data
print "no. of Chris training emails:", sum(labels_train)
print "no. of Sara training emails:", len(labels_train)-sum(labels_train)
return features_train_transformed, features_test_transformed, labels_train, labels_test
def predict(classifier_type="tree",selection="Univariate", f="1"):
if (f=="1"):
kc_fn = "GS_pickles\kmeans_Genes_87_1x_v3.pkl"
p = 1
BIG_C = 0.001
if (f=="2"):
kc_fn = "GS_pickles\kmeans_Genes_433_50x_v2.pkl"
p = 5
BIG_C = 0.1
if (f=="3"):
kc_fn = "GS_pickles\kmeans_Genes_2163_20x_v1.pkl"
p = 25
BIG_C = 2
dump_data = False
kernel_type = "linear"
(data_matrix, features, samples) = readData()
x = data_matrix.data
y = data_matrix.target
target_names = data_matrix.target_names
x_indices = np.arange(x.shape[-1])
(m,n) = x.shape
test = joblib.load("GS_pickles\imputed_test_data.pkl")
test_x = np.array(test)
(i,j) = test_x.shape
print "Training matrix shape: %s,%s" %(m,n)
print "Test matrix shape: %s,%s" %(i,j)
trimmed_x = []
trimmed_test_x = []
if (selection=="Univariate"):
selector = SelectPercentile(f_classif, percentile=p)
selector.fit(x, y)
# Trimming the matrix, now should contain x% of the 8650 features
trimmed_x = selector.transform(x)
trimmed_test_x = selector.transform(test_x)
if (selection=="kclusters"):
kcluster_flist = joblib.load(kc_fn)
trimmed_x = np.take(x, kcluster_flist, axis=1)
trimmed_test_x = np.take(test_x, kcluster_flist, axis=1)
n_samples, n_features = trimmed_x.shape
# Linear SVM classifier
if (classifier_type=="SVM"):
clf = svm.SVC(kernel=kernel_type, degree=3, probability=True)
# Gaussian Naive Bayes classifier
if (classifier_type=="NB"):
clf = GaussianNB()
clf.fit(trimmed_x,y)
result = clf.predict(trimmed_test_x)
return result
def eval(ds, testNum, p, splitProportion=0.2):
#testNum=1
#splitProportion=0.2
allFeaturesF1=[]
allFeaturesRecall=[]
allFeaturesPrecision=[]
featureSelctedF1=[]
featureSelctedRecall = []
featureSelctedPrecision = []
for _ in range(testNum):
tstdata, trndata = ds.splitWithProportion( splitProportion )
X, Y = labanUtil.fromDStoXY(trndata)
X_test, Y_test = labanUtil.fromDStoXY(tstdata)
#localF1s = []
#localRecalls = []
#localPercisions = []
for y, y_test in zip(Y, Y_test):
if all(v == 0 for v in y):
continue
#clf = LinearSVC()#fit_intercept=True, C=p)
#clf.sparsify()
#clf = RandomForestClassifier()#criterion='entropy')
#clf = tree.DecisionTreeClassifier()#max_depth=p)
clf = AdaBoostClassifier()
#clf = GradientBoostingClassifier()#, learning_rate=lr)
#clf = ExtraTreesClassifier(n_estimators=p)
#svc = LinearSVC()
#selector = RFE(estimator=svc, n_features_to_select=p*19, step=0.2)
selector = SelectPercentile(chooser, percentile=p)
selector.fit(X, y)
name = str(clf).split()[0].split('(')[0]
clf.fit(selector.transform(X), y)
pred = clf.predict(selector.transform(X_test))
featureSelctedF1.append(metrics.f1_score(y_test, pred))
featureSelctedRecall.append(metrics.recall_score(y_test, pred))
featureSelctedPrecision.append(metrics.precision_score(y_test, pred))
clf.fit(X, y)
pred = clf.predict(X_test)
allFeaturesF1.append(metrics.f1_score(y_test, pred))
allFeaturesRecall.append(metrics.recall_score(y_test, pred))
allFeaturesPrecision.append(metrics.precision_score(y_test, pred))
return np.mean(allFeaturesF1), np.mean(featureSelctedF1), \
np.mean(allFeaturesRecall), np.mean(featureSelctedRecall), \
np.mean(allFeaturesPrecision), np.mean(featureSelctedPrecision), \
name
def preprocess(words_file = "../tools/word_data.pkl", authors_file="../tools/email_authors.pkl"):
"""
this function takes a pre-made list of email texts (by default word_data.pkl)
and the corresponding authors (by default email_authors.pkl) and performs
a number of preprocessing steps:
-- splits into training/testing sets (10% testing)
-- vectorizes into tfidf matrix
-- selects/keeps most helpful features
after this, the feaures and labels are put into numpy arrays, which play nice with sklearn functions
4 objects are returned:
-- training/testing features
-- training/testing labels
"""
### the words (features) and authors (labels), already largely preprocessed
### this preprocessing will be repeated in the text learning mini-project
# read a vector of documents from file(decoded)
authors_file_handler = open(authors_file, "r")
authors = pickle.load(authors_file_handler)
authors_file_handler.close()
# read a vector of labels/authors from file(decoded)
words_file_handler = open(words_file, "r")
word_data = cPickle.load(words_file_handler)
words_file_handler.close()
### test_size is the percentage of events assigned to the test set
### (remainder go into training)
features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(word_data, authors, test_size=0.1, random_state=42)
# features_train,features_test is a vector of sentences
### text vectorization--go from strings to lists of numbers
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5,
stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test) # no fitting here. So the idf is the one calculated initially
# returns sparse matrix(N*M) where N = each document/sample, M gives tf*invdf weightage of current feature word in document.
### feature selection, because text is super high dimensional and
### can be really computationally chewy as a result
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(features_train_transformed, labels_train) # select top k% best features using univariate statistical tests
features_train_transformed = selector.transform(features_train_transformed).toarray() # select the columns based on the stats test
features_test_transformed = selector.transform(features_test_transformed).toarray() # do as above
### info on the data
#print "no. of Chris training emails:", sum(labels_train)
#print "no. of Sara training emails:", len(labels_train)-sum(labels_train)
return features_train_transformed, features_test_transformed, labels_train, labels_test
def reduce(self,percent_taken):
#fits classifier chi2 for non-negative X, otherwise F-value (ANOVA)
try:
fited=SelectPercentile(chi2, percentile=percent_taken).fit(self.train_set, self.Y)
except:
fited=SelectPercentile(f_classif, percentile=percent_taken).fit(self.train_set, self.Y)
self.fitted_reductor=fited
self.train_set = fited.transform(self.train_set)
self.test_set = fited.transform(self.test_set)
print 'number of featute(s) selected: {0}\n'.format(len(self.test_set[0]))
def make_train_test(df_train, df_test):
vectorizer = CountVectorizer()
X_train = vectorizer.fit_transform(df_train['Phrase'].values)
Y_train = df_train['Sentiment'].values
X_test = vectorizer.transform(df_test['Phrase'].values)
selector = SelectPercentile(f_classif, percentile=50)
selector.fit(X_train, Y_train)
features_train_transformed = selector.transform(X_train)
features_test_transformed = selector.transform(X_test)
return features_train_transformed, Y_train, features_test_transformed
def preprocess_4(article_file, lable_file):
# article_file = "pkl/2013_article.pkl"
# lable_file = "pkl/2013_lable.pkl"
features = pickle.load(open(article_file))
features = np.array(features)
# transform non-numerical labels (as long as they are hashable and comparable) to numerical labels
lables = pickle.load(open(lable_file))
le = preprocessing.LabelEncoder()
le.fit(lables)
lables = le.transform(lables)
### test_size is the percentage of events assigned to the test set (remainder go into training)
features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(features, lables, test_size=0.1, random_state=42)
# print features_train.shape
# print features_test[0]
# print features_test.shape
### text vectorization--go from strings to lists of numbers
t0 = time()
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5, min_df=1,
stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
# print "features_train_transformed is {}".format(features_train_transformed.shape)
# print "features_test_transformed is {}".format(features_test_transformed.shape)
# print "vectorizer time:", round(time()-t0, 3), "s"
# print len(vectorizer.get_feature_names())
### feature selection, because text is super high dimensional and
### can be really computationally chewy as a result
t0 = time()
selector = SelectPercentile(f_classif, percentile=30)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
# print "features_train_transformed is {}".format(features_train_transformed.shape)
# print "features_test_transformed is {}".format(features_test_transformed.shape)
# print "selector time:", round(time()-t0, 3), "s"
# print len(vectorizer.get_feature_names())
# print vectorizer.get_feature_names()[0:-10]
# print len(selector.scores_)
return features_train_transformed, features_test_transformed, labels_train, labels_test
def main():
#set the timer
start = time.time()
#load the data
trainX = np.load('trainX.npy')
testX = np.load('testX.npy')
trainY = np.load('trainY.npy')
testY = np.load('testY.npy')
print('\n!!! Data Loading Completed !!!\n')
#get the 1st digit zero and plot it
zero = trainX[14].reshape(28, 28)
plt.imshow(zero, cmap=cm.Greys_r)
plt.savefig("original"+str(trainY[14])+".png")
#plt.show()
#apply kpca
kpca = KernelPCA(kernel='rbf', gamma=1, fit_inverse_transform=True)
kpca.fit(trainX[0:3000])
trainX_kpca = kpca.transform(trainX)
testX_kpca = kpca.transform(testX)
#do inverse transform and plot the result
orig = kpca.inverse_transform(trainX_kpca)
img = orig[14].reshape(28, 28)
plt.imshow(img, cmap=cm.Greys_r)
plt.savefig("reconstructed"+str(trainY[14])+".png")
#plt.show()
selector = SelectPercentile(f_classif, percentile=5)
selector.fit(trainX_kpca, trainY)
trainX = selector.transform(trainX_kpca)
testX = selector.transform(testX_kpca)
#fit a classifier
parameters = {'n_neighbors' : list(np.arange(15)+1)}
clf = GridSearchCV(KNeighborsClassifier(weights='distance', n_jobs=-1), parameters)
clf.fit(trainX, trainY)
pred = clf.predict(testX)
print accuracy_score(testY, pred)
print confusion_matrix(testY, pred)
#print(clf.best_params_)
print('total : %d, correct : %d, incorrect : %d\n' %(len(pred), np.sum(pred == testY), np.sum(pred != testY)))
print('Test Time : %f Minutes\n' %((time.time()-start)/60))
def trainingPreprocess(words_file, authors_file):
"""
this function takes a pre-made list of email texts (by default word_data.pkl)
and the corresponding authors (by default email_authors.pkl) and performs
a number of preprocessing steps:
-- splits into training/testing sets (10% testing)
-- vectorizes into tfidf matrix
-- selects/keeps most helpful features
after this, the feaures and labels are put into numpy arrays, which play nice with sklearn functions
6 objects are returned:
-- training/testing features
-- training/testing labels
-- a fitted vectorizer
-- a fitted selector
"""
### the words (features) and authors (labels), already largely preprocessed
### this preprocessing will be repeated in the text learning mini-project
authors_file_handler = open(authors_file, "r")
authors = pickle.load(authors_file_handler)
authors_file_handler.close()
words_file_handler = open(words_file, "r")
word_data = cPickle.load(words_file_handler)
words_file_handler.close()
### test_size is the percentage of events assigned to the test set
### (remainder go into training)
features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(word_data, authors, test_size=0.1, random_state=42)
### text vectorization--go from strings to lists of numbers
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5,
stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
### feature selection
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
return features_train_transformed, features_test_transformed, labels_train, labels_test, vectorizer, selector
def selectFeatures(features, labels, features_list):
'''
Select features according to the 20th percentile of the highest scores.
Return a list of features selected and a dataframe showing the ranking
of each feature related to their p values
features: numpy array with the features to be used to test sklearn models
labels: numpy array with the real output
features_list: a list of names of each feature
'''
#feature selection
selector = SelectPercentile(f_classif, percentile=20)
selector.fit(features, labels)
features_transformed = selector.transform(features)
#filter names to be returned
l_rtn = [x for x, t in zip(features_list,
list(selector.get_support())) if t]
# pd.DataFrame(features_transformed, columns = l_labels2).head()
#calculate scores
scores = -np.log10(selector.pvalues_)
scores /= scores.max()
df_rtn = pd.DataFrame(pd.Series(dict(zip(features_list,scores))))
df_rtn.columns = ["pValue_Max"]
df_rtn = df_rtn.sort("pValue_Max", ascending=False)
# df_rtn["different_from_zero"]=((df!=0).sum()*1./df.shape[0])
return l_rtn, df_rtn
def preprocess(article_file, lable_file, k):
features = pickle.load(open(article_file))
features = np.array(features)
# transform non-numerical labels (as long as they are hashable and comparable) to numerical labels
lables = pickle.load(open(lable_file))
le = preprocessing.LabelEncoder()
le.fit(lables)
lables = le.transform(lables)
# print le.inverse_transform([0])
### text vectorization--go from strings to lists of numbers
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5, min_df=1,
stop_words='english')
features_train_transformed = vectorizer.fit_transform(features)
# selector : SelectPercentile
selector = SelectPercentile(f_classif, percentile=k)
selector.fit(features_train_transformed, lables)
# selector : chi2
# selector = SelectPercentile(score_func=chi2)
# selector.fit(features_train_transformed, lables)
features_train_transformed = selector.transform(features_train_transformed).toarray()
return features_train_transformed, lables, vectorizer, selector, le, features
def main():
main_data = pd.read_csv('../data/train.csv', index_col='ID')
output = []
for x in main_data.columns:
output.append({
'variable': x,
'variance': main_data.ix[:, x].var(),
'corr_w_target': round(main_data.ix[:, x].corr(main_data.TARGET), 4),
'abs_corr': abs(round(main_data.ix[:, x].corr(main_data.TARGET), 4))}
)
# print csv for later in the presentation docs
variable_selector = pd.DataFrame(output)
variable_selector = variable_selector.set_index('variable')
variable_selector = variable_selector.drop('TARGET')
variable_selector.sort_values('abs_corr', ascending=False).to_csv('../presentationDocs/corrs.csv')
selector = SelectPercentile(f_classif, percentile=25)
subset = pd.DataFrame(selector.fit_transform(main_data.drop('TARGET', axis=1), main_data['TARGET']))
subset.to_csv('../data/main_data.csv', index=False)
main_data[['TARGET']].to_csv('../data/target.csv', cols=['TARGET'], index=False)
# print transformed test data to csv
test_data = pd.read_csv('../data/test.csv', index_col='ID')
test_data = pd.DataFrame(selector.transform(test_data), index=test_data.index)
test_data.to_csv('../data/test_transform.csv', index=True, index_label='ID')
def select_features(X,y):
selector = SelectPercentile(f_classif, percentile=10)
print "fit selector"
selector.fit(X, y)
print "transform features"
X = selector.transform(X)
return X,selector
def train_type_model():
globals.read_configuration('config.cfg')
parser = globals.get_parser()
scorer_globals.init()
datasets = ["webquestions_split_train", ]
parameters = translator.TranslatorParameters()
parameters.require_relation_match = False
parameters.restrict_answer_type = False
feature_extractor = FeatureExtractor(False, False, n_gram_types_features=True)
features = []
labels = []
for dataset in datasets:
queries = get_evaluated_queries(dataset, True, parameters)
for index, query in enumerate(queries):
tokens = [token.lemma for token in parser.parse(query.utterance).tokens]
n_grams = get_grams_feats(tokens)
answer_entities = [mid for answer in query.target_result
for mid in KBEntity.get_entityid_by_name(answer, keep_most_triples=True)]
correct_notable_types = set(filter(lambda x: x,
[KBEntity.get_notable_type(entity_mid)
for entity_mid in answer_entities]))
other_notable_types = set()
for candidate in query.eval_candidates:
entities = [mid for entity_name in candidate.prediction
for mid in KBEntity.get_entityid_by_name(entity_name, keep_most_triples=True)]
other_notable_types.update(set([KBEntity.get_notable_type(entity_mid) for entity_mid in entities]))
incorrect_notable_types = other_notable_types.difference(correct_notable_types)
for type in correct_notable_types.union(incorrect_notable_types):
if type in correct_notable_types:
labels.append(1)
else:
labels.append(0)
features.append(feature_extractor.extract_ngram_features(n_grams, [type, ], "type"))
with open("type_model_data.pickle", 'wb') as out:
pickle.dump((features, labels), out)
label_encoder = LabelEncoder()
labels = label_encoder.fit_transform(labels)
vec = DictVectorizer(sparse=True)
X = vec.fit_transform(features)
feature_selector = SelectPercentile(chi2, percentile=5).fit(X, labels)
vec.restrict(feature_selector.get_support())
X = feature_selector.transform(X)
type_scorer = SGDClassifier(loss='log', class_weight='auto',
n_iter=1000,
alpha=1.0,
random_state=999,
verbose=5)
type_scorer.fit(X, labels)
with open("type-model.pickle", 'wb') as out:
pickle.dump((vec, type_scorer), out)
def preprocess(words_file="../tools/word_data.pkl", authors_file="../tools/email_authors.pkl"):
"""
Take a pre-made list of email texts (by default word_data.pkl) and the corresponding authors (by default email_authors.pkl) and preprocesses them.
Preprocessor steps:
- split into training/testing sets (10% testing)
- vectorize into tfidf matrix
- select/keep most helpful features
After this, the feaures and labels are put into numpy arrays, which play nice with sklearn functions
A tfidf matrix is defined as TF(t)*IDF(t) where
TF(t) = (Number of times term t appears in a document) / (Total number of terms in the document).
IDF(t) = log_e(Total number of documents / Number of documents with term t in it).
4 objects are returned:
-- training/testing features
-- training/testing labels
"""
# the words (features) and authors (labels), already largely preprocessed this preprocessing will be repeated in the text learning mini-project
print('words_file = {}'.format(words_file))
word_data = pickle.load(open(words_file, "rb"))
authors = pickle.load(open(authors_file, "rb"))
# test_size is the percentage of events assigned to the test set (remainder go into training)
features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(word_data, authors, test_size=0.1, random_state=42)
# text vectorization--go from strings to lists of numbers
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5, stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
# feature selection, because text is super high dimensional and can be really computationally chewy as a result
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
# info on the data
print("no. of Chris training emails:", sum(labels_train))
print("no. of Sara training emails:", len(labels_train) - sum(labels_train))
return numpy.array(features_train_transformed), numpy.array(features_test_transformed), numpy.array(labels_train), numpy.array(labels_test)
def feature_transform(features_train, features_test, top_percent=1):
""" Function to apply Bag of Words feature creator with TfIdf statistic
normalisation. The input is train and test text, and optional parameter
'top_percent' which shows how many percent of super high dimensional
text feature space is to return (defaul is 1%).
The output is the transformed train and test feature vectors suitable
to use with sklearn classifiers.
"""
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5,
stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
### Feature selection, because text is super high dimensional
selector = SelectPercentile(f_classif, percentile=top_percent)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
return features_train_transformed, features_test_transformed
def univariate_feature_selection(dataset, features):
# load the dataset
spreadsheet = Spreadsheet('../../Downloads/ip/project data.xlsx')
data = Data(spreadsheet)
targets = data.targets
X = dataset
y = data.targets
###############################################################################
plt.figure(1)
plt.clf()
X_indices = np.arange(X.shape[-1])
###############################################################################
# Univariate feature selection with F-test for feature scoring
# We use the default selection function: the 10% most significant features
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(X, y)
scores = -np.log10(selector.pvalues_)
scores /= scores.max()
plt.bar(X_indices - .45, scores, width=.2,
label=r'Univariate score ($-Log(p_{value})$)', color='g')
###############################################################################
# Compare to the weights of an SVM
clf = svm.SVC(kernel='linear')
clf.fit(X, y)
svm_weights = (clf.coef_ ** 2).sum(axis=0)
svm_weights /= svm_weights.max()
plt.bar(X_indices - .25, svm_weights, width=.2, label='SVM weight', color='r')
clf_selected = svm.SVC(kernel='linear')
clf_selected.fit(selector.transform(X), y)
svm_weights_selected = (clf_selected.coef_ ** 2).sum(axis=0)
svm_weights_selected /= svm_weights_selected.max()
plt.bar(X_indices[selector.get_support()] - .05, svm_weights_selected,
width=.2, label='SVM weights after selection', color='b')
x = np.arange(0, len(features))
plt.title("Comparing feature selection")
plt.xlabel('Feature number')
plt.xticks(x, features, rotation=45)
plt.yticks(())
#plt.axis('tight')
plt.legend(loc='upper right')
plt.show()
def preprocess(X,y):
### test_size is the percentage of events assigned to the test set
### (remainder go into training)
features_train, features_test, labels_train, labels_test = model_selection.train_test_split(X, y, test_size=0.2, random_state=42)
### text vectorization--go from strings to lists of numbers
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5, stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
joblib.dump(vectorizer, 'vectorizer_intent.pkl')
### feature selection, because text is super high dimensional and
### can be really computationally chewy as a result
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(features_train_transformed, labels_train)
joblib.dump(selector, 'selector_intent.pkl')
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
return features_train_transformed, features_test_transformed, labels_train, labels_test
def Preprocess(words_file="/home/mohamed/python/sherlok-tools/helpers/word_data.pkl", labels_file="/home/mohamed/python/sherlok-tools/helpers/label_data.pkl"):
"""
this function takes a pre-made list of data texts (by default word_data.pkl)
and the corresponding labels (by default label_data.pkl) and performs
a number of preprocessing steps:
-- splits into training/testing sets (10% testing)
-- vectorizes into tfidf matrix
-- selects/keeps most helpful features
after this, the feaures and labels are put into numpy arrays, which play nice with sklearn functions
4 objects are returned:
-- training/testing features
-- training/testing labels
"""
### the words (features) and labels (positive or negative)
word_data = pickle.load( open(words_file, "r"))
labels = pickle.load( open(labels_file, "r") )
### test_size is the percentage of events assigned to the test set (remainder go into training)
features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(word_data, labels, test_size=0.1, random_state=42)
### text vectorization--go from strings to lists of numbers
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5, encoding='windows-1256')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
### feature selection, because text is super high dimensional and
### can be really computationally chewy as a result
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
### info on the data
print "no. of positive training files:", sum(labels_train)
print "no. of negative training files:", len(labels_train)-sum(labels_train)
return features_train_transformed, features_test_transformed, labels_train, labels_test
def preprocess(words_file = "../data/data.pkl", authors_file="../data/datalabels.pkl"):
authors_file_handler = open(authors_file, "r")
authors = pickle.load(authors_file_handler)
authors_file_handler.close()
words_file_handler = open(words_file, "r")
word_data = cPickle.load(words_file_handler)
words_file_handler.close()
features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(word_data, authors, test_size=0.1, random_state=42)
features_train_vect = features_train
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5,
stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
selector = SelectPercentile(f_classif, percentile=1)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
return features_train_vect , features_train_transformed, features_test_transformed, labels_train, labels_test
def convertText(trainData, trainLabel, testData, testLabel, reduceDimensionality=0):
'''
trainData: training data
trainLabel: training labels
testData: test data
testLabel: test labels
return numerical arrays of data from text vectors
'''
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5, stop_words='english')
trainDataTransformed = vectorizer.fit_transform(trainData).toarray()
testDataTransformed = vectorizer.transform(testData).toarray()
if reduceDimensionality:
selector = SelectPercentile(f_classif, percentile=0.10)
selector.fit(trainDataTransformed, trainLabel)
trainDataTransformed = selector.transform(trainDataTransformed).toarray()
testDataTransformed = selector.transform(testDataTransformed).toarray()
return trainDataTransformed, trainLabel, testDataTransformed, testLabel
def eval(ds, clf, splitProportion=0.2, p=4):
#splitProportion = 0.2
tstdata, trndata = ds.splitWithProportion( splitProportion )
X, Y = labanUtil.fromDStoXY(trndata)
X_test, Y_test = labanUtil.fromDStoXY(tstdata)
f1s=[]
ps =[]
rs=[]
for i, (y, y_test) in enumerate(zip(Y, Y_test)):
if all(v == 0 for v in y):
continue
selector = SelectPercentile(chooser, percentile=p)
selector.fit(X, y)
name = str(clf).split()[0].split('(')[0]
clf.fit(selector.transform(X), y)
pred = clf.predict(selector.transform(X_test))
f1 = metrics.f1_score(y_test, pred)
f1s.append(f1)
ps.append(metrics.precision_score(y_test, pred))
rs.append(metrics.recall_score(y_test, pred))
return f1s, ps, rs
class AnovaPercentileStep(SklearnStep):
def __init__(self, percentile):
super(AnovaPercentileStep, self).__init__()
self._percentile = percentile
def fit_transform(self):
self._model = SelectPercentile(f_classif, self._percentile)
x, y = load_svmlight(self.input_path)
x = self._model.fit_transform(x, y)
save_svmlight(x, y, self._output_path)
def transform(self, x=None):
if x is None:
x, y = load_svmlight(self._test_input_path)
x = self._model.transform(x)
save_svmlight(x, y, self._test_output_path)
else:
transformed_x = self._model.transform(x)
return transformed_x
def get_param(self):
return {'percentile': self._percentile}
def train(config, model_data, data, record):
model_class_name, percentile = model_data
model = instantiate_from_class_string(model_class_name)
try:
model.n_jobs = config['n_jobs']
except:
log.info('Cannot set n_jobs for this model...')
record['model'] = model_name(model)
record['parameters'] = model.get_params()
record['feats_percentile'] = percentile
train_x = data['train_x']
train_y = data['train_y']
test_x = data['test_x']
# estimate accuracy using cross-validation
model = make_pipeline(SelectPercentile(f_classif, percentile),
StandardScaler(), model)
scores = cross_validation.cross_val_score(model, train_x,
train_y, cv=5,
scoring='accuracy')
record['mean_acc'] = scores.mean()
# predict on the test set
fn = SelectPercentile(f_classif, percentile).fit(train_x, train_y)
train_x = fn.transform(train_x)
test_x = fn.transform(test_x)
scaler = StandardScaler().fit(train_x)
train_x = scaler.transform(train_x)
test_x = scaler.transform(test_x)
model.fit(train_x, train_y)
ids = data['test_ids']
preds = model.predict(test_x)
record['test_preds'] = [(id_, pred) for id_, pred in zip(ids, preds)]
class FeatureSelection:
"""
特征选择
percentile：选取特征的百分比
"""
def __init__(self,percentile=70):
self.percentile=percentile
def fit(self,x,y):
self.sepChi=SelectPercentile(score_func=chi2,percentile=self.percentile)#使用卡方
self.sepChi.fit(x,y)
def transform(self,x,y):
return (self.sepChi.transform(x),y)
def preprocess_input(feature_test,words_file="/home/mohamed/python/sherlok-tools/helpers/word_data.pkl", labels_file="/home/mohamed/python/sherlok-tools/helpers/label_data.pkl"):
word_data = pickle.load( open(words_file, "r"))
labels = pickle.load( open(labels_file, "r") )
### test_size is the percentage of events assigned to the test set (remainder go into training)
### split training & testing
features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(word_data, labels, test_size=0.0, random_state=42)
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5, encoding='windows-1256')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(feature_test)
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
features_test_transformed = selector.transform(features_test_transformed).toarray()
### info on the data
print ("no. of positive training files:", sum(labels_train))
print ("no. of negative training files:", len(labels_train)-sum(labels_train))
return features_train_transformed, features_test_transformed, labels_train
def cross_val_score(clf, data, target, k):
shuffle_arr = []
size = len(data)
for i in range(size):
shuffle_arr.append(i)
scores = []
for i in range(0, k):
#generate shuffled train and test dataset
data_train_raw = []
data_test_raw = []
target_train = []
target_test = []
# seperate shuffled train and test dataset
random.shuffle(shuffle_arr)
shuffle_train = shuffle_arr[:size - size/k]
shuffle_test = shuffle_arr[size-size/k :]
for j in shuffle_train:
data_train_raw.append(data_total[j])
target_train.append(target[j])
for r in shuffle_test:
data_test_raw.append(data_total[r])
target_test.append(target[r])
data_train = data_process(data_train_raw)
data_test = data_process(data_test_raw)
# transform array of string to counts
count_vect = CountVectorizer()
X_train_counts = count_vect.fit_transform(data_train)
# transform counts to frequencies
tf_transformer = TfidfTransformer(use_idf=False).fit(X_train_counts)
X_train_tf = tf_transformer.transform(X_train_counts)
# feature selection
select = SelectPercentile(chi2, percentile = 10)
X_train_fs = select.fit_transform(X_train_tf, target_train)
# train the model
clf_train = clf.fit(X_train_fs, target_train)
# test the model
X_new_counts = count_vect.transform(data_test)
X_new_tfidf = tf_transformer.transform(X_new_counts)
X_new_fs = select.transform(X_new_tfidf)
test_result = clf_train.predict(X_new_fs)
scores.append(GetPrecisionRecallF1(test_result, target_test))
#clf_score = clf_train.score(X_new_fs, target_test)
#scores.append(clf_score)
return scores
def get_combined_separate_fsets(feature_sets, fs_fn='pct', ptile=10, nFeatures=5, score_fn=f_classif):
df_lst = []
for fset_name, df in feature_sets.items():
X_train = df[df.partition == 'train'].drop(['partition', 'fatality_ind'], axis=1)
y_train = df[df.partition == 'train'].fatality_ind
df_X = df.drop(['partition', 'fatality_ind'], axis=1)
if fs_fn == 'pct':
featureSelector = SelectPercentile(score_func=score_fn, percentile=ptile)
else:
featureSelector = SelectKBest(score_func=score_fn, k=nFeatures)
featureSelector.fit(X_train, y_train)
fs = featureSelector.transform(df.drop(['partition', 'fatality_ind'], axis=1))
cols_fs = df_X.columns[list(featureSelector.get_support(indices=True))]
cols_fs_ref = [fset_name + ' ' + c for c in cols_fs]
df_fs = pd.DataFrame(fs, index=df_X.index, columns=cols_fs_ref)
df_lst.append(df_fs)
df_comb = df[['partition', 'fatality_ind']].join(pd.concat(df_lst, axis=1))
return df_comb
features_train = pre_process(features_train)
features_test = pre_process(features_test)
vectorizer = TfidfVectorizer(sublinear_tf=True, max_df=0.5,
stop_words='english')
features_train_transformed = vectorizer.fit_transform(features_train)
features_test_transformed = vectorizer.transform(features_test)
#print features_train_transformed
#print features_train_transformed.shape
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(features_train_transformed, labels_train)
features_train_transformed = selector.transform(features_train_transformed).toarray()
'''print '--------------------------------------------------------------------'
print features_train_transformed
print features_train_transformed.shape
print labels_train.shape'''
features_test_transformed = selector.transform(features_test_transformed).toarray()
'''print features_test_transformed.shape
print labels_test.shape
print type(features_train_transformed)
print type(labels_train)'''
clf = GaussianNB()
t0 = time()
clf.fit(features_train_transformed, labels_train)
print "training time:", round(time()-t0, 3), "s"
from sklearn.feature_selection import SelectPercentile
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
cancer = load_breast_cancer()
rng = np.random.RandomState(42)
noise = rng.normal(size=(len(cancer.data), 50))
X_w_noise = np.hstack([cancer.data, noise])
X_train, X_test, y_train, y_test = train_test_split(
X_w_noise, cancer.target, random_state=0, test_size=.5)
select = SelectPercentile(percentile=50)
select.fit(X_train, y_train)
X_train_selected = select.transform(X_train)
mask = select.get_support()
print(mask)
plt.matshow(mask.reshape(1, -1), cmap='gray_r')
plt.xlabel('Sample index')
plt.show()
X_test_selected = select.transform(X_test)
lr = LogisticRegression()
lr.fit(X_train, y_train)
print('Score with all features: {:.3f}'.format(lr.score(X_test, y_test)))
lr.fit(X_train_selected, y_train)
print('Score with selected features: {:.3f}'.format(lr.score(X_test_selected, y_test)))
features_cv_transformed=selector.transform(features_cv_transformed)
selected_word_indices = selector.get_support(indices=True)
vocab = vectorizer.get_feature_names()
trimmed_vocab = [vocab[i] for i in selected_word_indices]
print[trimmed_vocab]
'''
from sklearn.feature_selection import SelectPercentile, f_classif
from sklearn.feature_extraction.text import TfidfVectorizer
vectorizer = TfidfVectorizer(sublinear_tf=True,
max_df=0.5,
stop_words="english")
features_transformed = vectorizer.fit_transform(word_data)
features_transformed = features_transformed.toarray()
selector = SelectPercentile(f_classif, percentile=20)
selector.fit(features_transformed, sentiid)
features_transformed = selector.transform(features_transformed)
##################################################################################################################################################
from sklearn.svm import SVC
from sklearn.datasets import load_digits
from sklearn.model_selection import learning_curve
from sklearn.model_selection import ShuffleSplit
def plot_learning_curve(estimator,
title,
X,
y,
ylim=None,
cv=None,
n_jobs=1,
train_sizes=np.linspace(.1, 1.0, 5)):
tokenizer=stem_comment,
max_features=5000)
dtm = vect.fit_transform(Xtrain)
words = vect.get_feature_names()
print('dtm matrix')
from sklearn.feature_selection import SelectPercentile, mutual_info_classif
selector = SelectPercentile(mutual_info_classif, percentile=20)
dtm_reduced = selector.fit_transform(dtm, ytrain)
selector_scores = selector.scores_
print('selected')
dtm_test = vect.transform(Xtest)
dtm_selected = selector.transform(dtm_test)
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report
from sklearn.metrics import roc_curve, auc, confusion_matrix
from sklearn.preprocessing import label_binarize
model_random_forest = RandomForestClassifier()
model_random_forest.fit(dtm_reduced, ytrain)
prob_pred = model_random_forest.predict_proba(dtm_selected)
pred = model_random_forest.predict(dtm_selected)
cm_random_forest = confusion_matrix(ytest, pred)
cr_random_forest = classification_report(ytest, pred)
#%%
train = data[:train.shape[0]]
test = data[train.shape[0]:]
train_y = train['click']
cntv = CountVectorizer()
cntv.fit(train['user_tags'])
train_a = cntv.transform(train['user_tags'])
test_a = cntv.transform(test['user_tags'])
train_new = sparse.hstack(
(train_new, train_a), 'csr'
) #hstack ： 将矩阵按照列进行拼接，对应的列数必须相等，hstack(blocks, format=None, dtype=None)
test_new = sparse.hstack((test_new, test_a), 'csr')
SKB = SelectPercentile(chi2, percentile=95).fit(
train_new,
train_y) #区别：SelectKBest选择排名排在前n个的变量 SelectPercentile 选择排名排在前n%的变量
train_new = SKB.transform(train_new)
test_new = SKB.transform(test_new)
'''
在稀疏矩阵存储格式中：
# - COO 格式在构建矩阵时比较高效
# - CSC 和 CSR 格式在乘法计算时比较高效
A.todense()
# 可以转化为普通矩阵：
'''
#%%
#统计特征构造
#adid统计特征，不同种类数量（已经创建了记录数的统计，现在是一个特征对应另外一个特征的种类）
## 由于adid是次样本层级的粒度，是聚集到点击率的层面所以是重要的特征，adid基本与广告信息表一一对应，我们象征性的选择广告id与其他挑选出来的id进行特征nunique统计
adid_nuq = [
'model', 'make', 'os', 'city', 'province', 'user_tags', 'app_id',
'carrier', 'nnt', 'devtype', 'app_cate_id', 'inner_slot_id'
def train_classifier_use_feature_selection(self):
# Get list of features
count_vect = CountVectorizer(stop_words=stopwords, min_df=3, max_df=0.90, ngram_range=_ngram_range)
X_CV = count_vect.fit_transform(docs_train)
# print number of unique words (n_features)
print ("Shape of train data is "+str(X_CV.shape))
# tfidf transformation###
tfidf_transformer = TfidfTransformer(use_idf=_use_idf)
X_tfidf = tfidf_transformer.fit_transform(X_CV)
#################
# feature selection
#################
selector = SelectPercentile(score_func=_score_func, percentile=_percentile)
print ("Fitting data with feature selection ...")
selector.fit(X_tfidf, y_train)
# get how many features are left after feature selection
X_features = selector.transform(X_tfidf)
print ("Shape of array after feature selection is "+str(X_features.shape))
clf = MultinomialNB(alpha=_alpha).fit(X_features, y_train)
# get the features which are selected and write to file
feature_boolean = selector.get_support(indices=False)
f = open(path_to_store_feature_selection_boolean_file,'w')
for fb in feature_boolean:
f.write(str(fb)+'\n')
f.close()
##################
# get cross validation score
##################
scores = cross_val_score(clf, X_features, y_train, cv=10, scoring='f1_weighted')
print ("Cross validation score: "+str(scores))
# Get average performance of classifier on training data using 10-fold CV, along with standard deviation
print("Cross validation accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
####################
#test clf on test data
####################
X_test_CV = count_vect.transform(docs_test)
print ("Shape of test data is "+str(X_test_CV.shape))
X_test_tfidf = tfidf_transformer.transform(X_test_CV)
# apply feature selection on test data too
X_test_selector = selector.transform(X_test_tfidf)
print ("Shape of array for test data after feature selection is "+str(X_test_selector.shape))
y_predicted = clf.predict(X_test_selector)
# print the mean accuracy on the given test data and labels
print ("Classifier score on test data is: %0.2f " % clf.score(X_test_selector,y_test))
print(metrics.classification_report(y_test, y_predicted))
cm = metrics.confusion_matrix(y_test, y_predicted)
print(cm)
return clf, count_vect
def benchmark(X_train, X_test, y5_train, y5_test, y3_train, y3_test, y2_train,
y2_test, exp_folder, ds_folder, perc_f):
config.logger.info('benchmark_text: ' + str(perc_f))
try:
subfolder = 'all/best_k/' + str(perc_f) + '/'
path = OUTPUT_FOLDER + exp_folder + ds_folder + 'benchmark/' + subfolder
#input_layer_neurons = len(X) + 1
#output_layer_neurons = 1
#hidden_nodes = np.math.ceil(len(X) / (2 * (input_layer_neurons + output_layer_neurons)))
#X_train_3, X_test_3, y_train_3, y_test_3 = train_test_split(X, y3, test_size=test_size, random_state=random_state)
#X_train_2, X_test_2, y_train_2, y_test_2 = train_test_split(X, y2, test_size=test_size, random_state=random_state)
# just to double check...
#assert np.all(X_train_5 == X_train_3)
#assert np.all(X_train_5 == X_train_2)
scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)
config.logger.debug('OK. feature selection')
# feature selection
best5 = SelectPercentile(f_regression, perc_f)
best3 = SelectPercentile(f_classif, perc_f)
best2 = SelectPercentile(f_classif, perc_f)
X_train_best5 = best5.fit_transform(X_train, y5_train)
X_test_best5 = best5.transform(X_test)
#feature_names = ['a', 'b', 'c', 'd', 'e']
#best_scores = best5.scores_
#best_features_ordered = [feature_names[i] for i in np.argsort(best5.scores_)[::-1]]
X_train_best3 = best3.fit_transform(X_train, y3_train)
X_test_best3 = best3.transform(X_test)
X_train_best2 = best2.fit_transform(X_train, y2_train)
X_test_best2 = best2.transform(X_test)
best_estimators = []
x_axis_2 = []
x_axis_3 = []
y_axis_2 = []
y_axis_3 = []
title = 'Webpage Text Features'
x_axis_label = 'Classifiers'
y_axis_label = 'F1-measure'
# --------------------------------------------------------------------------------------------------------------
# regression experiment
# --------------------------------------------------------------------------------------------------------------
config.logger.info('starting experiments regression (5-classes)')
with open(path + EXP_5_CLASSES_LABEL + '/log/results.txt',
"w") as file_log_regression:
file_log_regression.write(HEADER_CLASSIFICATION)
for estimator, hyperparam, grid_method in CONFIGS_REGRESSION:
out = []
label = estimator.__class__.__name__ + '.' + str(
perc_f) + '.' + EXP_5_CLASSES_LABEL
out, best_estimator = train_test_export_save_per_exp_type(
estimator, label, hyperparam, grid_method, X_train_best5,
X_test_best5, y5_train, y5_test, EXP_5_CLASSES_LABEL, 0,
out, file_log_regression, subfolder, exp_folder, ds_folder)
file_log_regression.flush()
# --------------------------------------------------------------------------------------------------------------
# classification experiment
# --------------------------------------------------------------------------------------------------------------
config.logger.info(
'starting experiments classification (2-classes and 3-classes)')
i = 1
for exp_type in (EXP_2_CLASSES_LABEL, EXP_3_CLASSES_LABEL):
with open(path + exp_type + '/log/results.txt',
"w") as file_log_classification:
file_log_classification.write(HEADER_CLASSIFICATION)
if exp_type == EXP_2_CLASSES_LABEL:
_X_train = X_train_best2
_X_test = X_test_best2
_y_train = y2_train
_y_test = y2_test
y_axis = y_axis_2
x_axis = x_axis_2
graph_file = 'graph.' + str(perc_f) + '.2-class.png'
threshold = THRESHOLD_LABEL_2class
elif exp_type == EXP_3_CLASSES_LABEL:
_X_train = X_train_best3
_X_test = X_test_best3
_y_train = y3_train
_y_test = y3_test
y_axis = y_axis_3
x_axis = x_axis_3
graph_file = 'graph.' + str(perc_f) + '.3-class.png'
threshold = THRESHOLD_LABEL_3class
else:
raise Exception('blah! error')
for estimator, hyperparam, grid_method in CONFIGS_CLASSIFICATION:
out = []
label = estimator.__class__.__name__ + '.' + str(
perc_f) + '.' + exp_type
out, best_estimator = train_test_export_save_per_exp_type(
estimator, label, hyperparam, grid_method, _X_train,
_X_test, _y_train, _y_test, exp_type, 0, out,
file_log_classification, subfolder, exp_folder,
ds_folder)
best_estimators.append(
(estimator.__class__.__name__, best_estimator))
i += 1
y_axis.extend(np.array(out)[:, 2])
x_axis.append(
best_estimator.__class__.__name__.replace(
'Classifier', ''))
#estimator_ensamble = VotingClassifier(estimators=best_estimators)
#hyperparam_ensamble = dict(voting=['hard', 'soft'], flatten_transform=[True, False])
#out = []
#out, best_estimator = train_test_export_save_per_exp_type(estimator_ensamble, estimator_ensamble.__class__.__name__,
# hyperparam_ensamble, SEARCH_METHOD_GRID,
# _X_train, _X_test, _y_train, _y_test, exp_type, 0,
# out, file_log_classification, subfolder, exp_folder, ds_folder)
#y_axis.extend(np.array(out)[:, 2])
#x_axis.append(best_estimator.__class__.__name__.replace('Classifier', ''))
config.logger.info(
'experiments classification done! exporting charts...')
export_chart_bar(x_axis, y_axis, graph_file, ds_folder,
exp_folder, perc_f, exp_type, title,
x_axis_label, y_axis_label, threshold)
config.logger.info('charts exported!')
file_log_classification.flush()
except Exception as e:
config.logger.error(repr(e))
raise
target = pickle.load(open("../generated/group.p", "r"))
device_id = pickle.load(open("../generated/device_id.p", "r"))
trainDevices = pd.read_csv("../../../data/gender_age_train.csv",
usecols=["device_id"])
indexes = pd.read_csv("../generated/raddarIndices.csv")
indexes = pd.merge(trainDevices,
indexes,
how="left",
on="device_id",
left_index=True).reset_index().drop(["index"], axis=1)
######################
# Feature Selection
######################
fs = SelectPercentile(chi2, percentile=23).fit(train, target)
train = fs.transform(train)
test = fs.transform(test)
##################
# Pre Processing
##################
targetEncoder = LabelEncoder()
target = targetEncoder.fit_transform(target)
target = np_utils.to_categorical(target)
##################
# Build Model
##################
def modelBuilder():
model = Sequential()
def main():
data_loc = sys.argv[1]
stemmer = PorterStemmer()
# Get data for all combos
train_stem_stop = load_files_correctly(os.path.join(data_loc, 'Training'),
stemmer=stemmer,
stop=True)
test_stem_stop = load_files_correctly(os.path.join(data_loc, 'Test'),
stemmer=stemmer,
stop=True)
train_no_stem_stop = load_files_correctly(os.path.join(
data_loc, 'Training'),
stemmer=None,
stop=True)
test_no_stem_stop = load_files_correctly(os.path.join(data_loc, 'Test'),
stemmer=None,
stop=True)
train_stem_no_stop = load_files_correctly(os.path.join(
data_loc, 'Training'),
stemmer=stemmer,
stop=False)
test_stem_no_stop = load_files_correctly(os.path.join(data_loc, 'Test'),
stemmer=stemmer,
stop=False)
train_no_stem_no_stop = load_files_correctly(os.path.join(
data_loc, 'Training'),
stemmer=None,
stop=False)
test_no_stem_no_stop = load_files_correctly(os.path.join(data_loc, 'Test'),
stemmer=None,
stop=False)
tfid_u_stem_stop = TfidfVectorizer(ngram_range=(1, 1),
decode_error='ignore')
tfid_u_stem_no_stop = TfidfVectorizer(ngram_range=(1, 1),
decode_error='ignore')
tfid_u_no_stem_stop = TfidfVectorizer(ngram_range=(1, 1),
decode_error='ignore')
tfid_u_no_stem_no_stop = TfidfVectorizer(ngram_range=(1, 1),
decode_error='ignore')
count_u_stem_stop = CountVectorizer(ngram_range=(1, 1),
decode_error='ignore')
count_u_stem_no_stop = CountVectorizer(ngram_range=(1, 1),
decode_error='ignore')
count_u_no_stem_stop = CountVectorizer(ngram_range=(1, 1),
decode_error='ignore')
count_u_no_stem_no_stop = CountVectorizer(ngram_range=(1, 1),
decode_error='ignore')
tfid_u_train_stem_stop = tfid_u_stem_stop.fit_transform(
train_stem_stop.data)
tfid_u_test_stem_stop = tfid_u_stem_stop.transform(test_stem_stop.data)
tfid_u_train_stem_no_stop = tfid_u_stem_no_stop.fit_transform(
train_stem_no_stop.data)
tfid_u_test_stem_no_stop = tfid_u_stem_no_stop.transform(
test_stem_no_stop.data)
tfid_u_train_no_stem_stop = tfid_u_no_stem_stop.fit_transform(
train_no_stem_stop.data)
tfid_u_test_no_stem_stop = tfid_u_no_stem_stop.transform(
test_no_stem_stop.data)
tfid_u_train_no_stem_no_stop = tfid_u_no_stem_no_stop.fit_transform(
train_no_stem_no_stop.data)
tfid_u_test_no_stem_no_stop = tfid_u_no_stem_no_stop.transform(
test_no_stem_no_stop.data)
count_u_train_stem_stop = count_u_stem_stop.fit_transform(
train_stem_stop.data)
count_u_test_stem_stop = count_u_stem_stop.transform(test_stem_stop.data)
count_u_train_stem_no_stop = count_u_stem_no_stop.fit_transform(
train_stem_no_stop.data)
count_u_test_stem_no_stop = count_u_stem_no_stop.transform(
test_stem_no_stop.data)
count_u_train_no_stem_stop = count_u_no_stem_stop.fit_transform(
train_no_stem_stop.data)
count_u_test_no_stem_stop = count_u_no_stem_stop.transform(
test_no_stem_stop.data)
count_u_train_no_stem_no_stop = count_u_no_stem_no_stop.fit_transform(
train_no_stem_no_stop.data)
count_u_test_no_stem_no_stop = count_u_no_stem_no_stop.transform(
test_no_stem_no_stop.data)
# Vectorize data
res = []
name = 'Naive Bayes alpha=0.01; unigram; countvectorizer; no stemmer; no stopper; no feature selection'
clf = MultinomialNB(alpha=0.01)
clf.fit(count_u_train_no_stem_no_stop, train_no_stem_no_stop.target)
pred = clf.predict(count_u_test_no_stem_no_stop)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_no_stem_no_stop.target
})
name = 'Naive Bayes alpha=0.01; unigram; tfidfvectorizer; no stemmer; no stopper; no feature selection'
clf = MultinomialNB(alpha=0.01)
clf.fit(tfid_u_train_no_stem_no_stop, train_no_stem_no_stop.target)
pred = clf.predict(tfid_u_test_no_stem_no_stop)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_no_stem_no_stop.target
})
name = 'Naive Bayes alpha=0.01 fit_prior=False; unigram; tfidfvectorizer; no stemmer; no stopper; no feature selection'
clf = MultinomialNB(alpha=0.01, fit_prior=False)
clf.fit(tfid_u_train_no_stem_no_stop, train_no_stem_no_stop.target)
pred = clf.predict(tfid_u_test_no_stem_no_stop)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_no_stem_no_stop.target
})
name = 'Naive Bayes alpha=0.01; unigram; countvectorizer; stemmer; no stopper; no feature selection'
clf = MultinomialNB(alpha=0.01)
clf.fit(count_u_train_stem_no_stop, train_stem_no_stop.target)
pred = clf.predict(count_u_test_stem_no_stop)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_stem_no_stop.target
})
name = 'Naive Bayes alpha=0.01; unigram; countvectorizer; no stemmer; stopper; no feature selection'
clf = MultinomialNB(alpha=0.01)
clf.fit(count_u_train_no_stem_stop, train_no_stem_stop.target)
pred = clf.predict(count_u_test_no_stem_stop)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_no_stem_stop.target
})
name = 'Naive Bayes alpha=0.01; unigram; countvectorizer; stemmer; stopper; no feature selection'
clf = MultinomialNB(alpha=0.01)
clf.fit(count_u_train_stem_stop, train_stem_stop.target)
pred = clf.predict(count_u_test_stem_stop)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_stem_stop.target
})
name = 'Naive Bayes alpha=0.01 fit_prior=False; unigram; tfidfvectorizer; no stemmer; stopper; no feature selection'
clf = MultinomialNB(alpha=0.01, fit_prior=False)
clf.fit(tfid_u_train_no_stem_stop, train_no_stem_stop.target)
pred = clf.predict(tfid_u_test_no_stem_stop)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_no_stem_stop.target
})
name = 'Naive Bayes alpha=0.01 fit_prior=False; unigram; tfidfvectorizer; stemmer; stopper; no feature selection'
clf = MultinomialNB(alpha=0.01, fit_prior=False)
clf.fit(tfid_u_train_stem_stop, train_stem_stop.target)
pred = clf.predict(tfid_u_test_stem_stop)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_stem_stop.target
})
name = 'Naive Bayes alpha=0.01; unigram; countvectorizer; no stemmer; stopper; feature selection - SelectPercentile=80'
ch2 = SelectPercentile(chi2, percentile=80)
X_train = ch2.fit_transform(count_u_train_no_stem_stop,
train_no_stem_stop.target)
X_test = ch2.transform(count_u_test_no_stem_stop)
clf = MultinomialNB(alpha=0.01)
clf.fit(X_train, train_no_stem_stop.target)
pred = clf.predict(X_test)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_no_stem_stop.target
})
name = 'Naive Bayes alpha=0.01 fit_prior=False; unigram; tfifdfvectorizer; stemmer; stopper; feature selection - SelectPercentile=80'
ch2 = SelectPercentile(chi2, percentile=80)
X_train = ch2.fit_transform(tfid_u_train_stem_stop, train_stem_stop.target)
X_test = ch2.transform(tfid_u_test_stem_stop)
clf = MultinomialNB(alpha=0.01, fit_prior=False)
clf.fit(X_train, train_stem_stop.target)
pred = clf.predict(X_test)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_stem_stop.target
})
name = 'Naive Bayes alpha=0.01; unigram; countvectorizer; no stemmer; stopper; feature selection - SelectFromModel threshold=39'
clf = MultinomialNB(alpha=0.01)
clf.fit(count_u_train_no_stem_stop, train_no_stem_stop.target)
model = SelectFromModel(clf, threshold=30, prefit=True)
X_new = model.transform(count_u_train_no_stem_stop)
X_new_test = model.transform(count_u_test_no_stem_stop)
clf = MultinomialNB(alpha=0.01)
clf.fit(X_new, train_no_stem_stop.target)
pred = clf.predict(X_new_test)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_no_stem_stop.target
})
name = 'Naive Bayes alpha=0.01 fit_prior=False; unigram; tfidfvectorizer; stemmer; stopper; feature selection - SelectFromModel threshold=30'
clf = MultinomialNB(alpha=0.01, fit_prior=False)
clf.fit(tfid_u_train_stem_stop, train_stem_stop.target)
model = SelectFromModel(clf, threshold=30, prefit=True)
X_new_train = model.transform(tfid_u_train_stem_stop)
X_new_test = model.transform(tfid_u_test_stem_stop)
clf = MultinomialNB(alpha=0.01, fit_prior=False)
clf.fit(X_new_train, train_stem_stop.target)
pred = clf.predict(X_new_test)
res.append({
'name': name,
'clf': clf,
'pred': pred,
'target': test_stem_stop.target
})
# This is the best model
best_model = name
rows = []
headers = [
'No.', 'Name', 'Precision', 'Recall', 'Precision/Recall', 'F1 Score'
]
for i, val in enumerate(res):
precision = metrics.precision_score(val['target'],
val['pred'],
average='macro')
recall = metrics.recall_score(val['target'],
val['pred'],
average='macro')
f1_score = metrics.f1_score(val['target'],
val['pred'],
average='macro')
row = [
i + 1, val['name'], precision, recall, precision / recall, f1_score
]
rows.append(row)
print('\nResults - ')
print tabulate(rows, headers, tablefmt='orgtbl')
print('\n')
print 'Best model is ->', best_model
def feature_select(self):
feature_select = SelectPercentile(chi2, percentile=95)
feature_select.fit(self.train_x, self.train_y)
train_csr = feature_select.transform(self.train_x)
predict_csr = feature_select.transform(self.test_x)
return train_csr, predict_csr
def process(self, df):
print('process:', self.name)
# 1). create strings based on text
train = df[df['type'] == 'train']
print(train.head())
test = df[df['type'] == 'test']
print(test.head())
print('train.shape:', train.shape)
n_train = train.shape[0]
print('n_train:', n_train)
n_test = test.shape[0]
print('n_test:', n_test)
# 2). fit a TfidfVectorizer on text
vec_text = TfidfVectorizer(analyzer='char', ngram_range=(1, 2),
max_df=0.8, min_df=2, sublinear_tf=True)
text_tfidf = vec_text.fit_transform(df['text'].tolist())
print('text Tfidf.shape:', text_tfidf.shape)
vocabulary = vec_text.vocabulary_
print('vocabulary size:%d' % len(vocabulary))
print('vocabulary list:')
count = 0
for k, v in vocabulary.items():
if count < 10:
print("%s\t%s" % (k, v))
count += 1
print("feature set nums: ", len(vocabulary))
feature_names = vec_text.get_feature_names()
ch2_precent = SelectPercentile(chi2, percentile=5)
ch2 = ch2_precent.fit(text_tfidf[:n_train], df.iloc[:n_train]['label'])
text_tfidf = ch2_precent.transform(text_tfidf)
features = [feature_names[i] for i in ch2.get_support(indices=True)]
feature_scores = [ch2.scores_[i] for i in ch2.get_support(indices=True)]
sorted_feature = sorted(zip(features, feature_scores), key=lambda x: x[1], reverse=True)
feature_output_file = config.output_dir + 'char_tfidf_feature.txt'
with open(feature_output_file, "w", encoding="utf-8") as f:
for id, item in enumerate(sorted_feature):
f.write("\t".join([str(id + 1), item[0], str(item[1])]) + "\n")
print("feature select done, new feature set num: ", len(feature_scores))
# save train and test into separate files
tfidf_train = text_tfidf[:n_train, :]
tfidf_train_feature_path = config.output_dir + "train.text.char.tfidf.pkl"
with open(tfidf_train_feature_path, "wb") as f:
pickle.dump(tfidf_train, f)
print('text tfidf features of training set saved in %s' % tfidf_train_feature_path)
tfidf_test = None
if n_test > 0:
# test set is available
tfidf_test = text_tfidf[n_train:, :]
tfidf_test_feature_path = config.output_dir + "test.text.char.tfidf.pkl"
with open(tfidf_test_feature_path, "wb") as f:
pickle.dump(tfidf_test, f)
print('text tfidf features of test set saved in %s' % tfidf_test_feature_path)
return tfidf_train.toarray(), tfidf_test.toarray(), train['label'].values
]
data_frame = data_frame.fillna(0)
# Store to my_dataset for easy export below.
my_dataset = data_frame.to_dict('index')
print("")
print("New data_dict:\n", my_dataset)
# Extract features and labels from dataset for local testing
data = featureFormat(my_dataset, features_list, sort_keys=True)
labels, features = targetFeatureSplit(data)
from sklearn.feature_selection import SelectPercentile, f_classif
selector = SelectPercentile(f_classif, percentile=50)
selector.fit(features, labels)
features_train = selector.transform(features)
features_test = selector.transform(features)
SelectPercentile_features = zip(selector.get_support(), features_list[1:], selector.scores_)
SelectPercentile_features = sorted(SelectPercentile_features, key=lambda x: x[2], reverse=True)
print ("(Features marked with 'True' are used in the final algorithm.):")
for feature in SelectPercentile_features:
print(feature)
# Task 4: Try a variety of classifiers
# Provided to give you a starting point. Try a variety of classifiers.
# from sklearn.naive_bayes import GaussianNB
# from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import GradientBoostingClassifier
# Task 5: Tune your classifier to achieve better than .3 precision and recall using our testing script.
clf = GradientBoostingClassifier(init=None,
def lightgbm_make_submission():
train, test = read_csv()
# x_train, x_test, y_train, y_test = make_train_set()
x_train, y_train = make_train_set(train)
# df = pd.merge(x_train, y_train, on='TERMINALNO')
# if PREDICT == False:
# df.corr().to_csv("./data/corr.csv")
# else:
# print(df.corr())
# del df
# gc.collect()
x_test, y_test = make_train_set(test)
y_train = y_train['Y']
# feature selection
sel = SelectPercentile(f_regression, 50)
x_train = sel.fit_transform(x_train, y_train)
x_test = sel.transform(x_test)
# print("**********************x_train*******************")
# print(x_train)
# print("**********************x_train end***************")
# print("**********************x_test********************")
# print(x_test.head())
# print("**********************x_test end****************")
train_x, valid_x, train_y, valid_y = train_test_split(x_train,
y_train,
test_size=0.2,
random_state=0)
params = {
# 'boosting': 'dart',
'learning_rate': 0.01,
'application': 'regression',
'max_depth': -1,
'num_leaves': 5,
'verbosity': -1,
'feature_fraction': 0.8,
'feature_fraction_seed': 9,
'bagging_fraction': 0.8,
'bagging_freq': 5,
'bagging_seed': 9,
'min_data_in_leaf': 6,
'min_sum_hessian_in_leaf': 11,
'metric': 'mae',
}
d_train = lgb.Dataset(train_x, label=train_y)
d_valid = lgb.Dataset(valid_x, label=valid_y)
watchlist = [d_train, d_valid]
model = lgb.train(params,
train_set=d_train,
num_boost_round=300,
valid_sets=watchlist,
verbose_eval=20)
# if PREDICT:
print(
"*******************************start predict***************************"
)
preds = model.predict(x_test)
y_test['Y'] = preds
print(y_test['Y'].var())
y_test.columns = ['TERMINALNO', 'Pred']
y_test.set_index('TERMINALNO', inplace=True)
y_test.to_csv(path_test_out,
columns=['Pred'],
index=True,
index_label=['Id'])
print(df.dtypes) #datatype of the columns
##instantiate the predictor variables & the target variable (with known classes encoded in numbers)
dataset = df.values #returns a numpy array format of the dataset
x = dataset[:,
0:7] #each column from 1 - 7 contains the predictor variables (x)
y = dataset[:,
7] #the last column 8 contains the target variable which is the "class" of the localisation site (y)
#(Optional) feature engineering: automatic feature selection to reduce dimensionality
#(I)Univeriate statistics method by SelectPercentile
select = SelectPercentile(
percentile=50
) #(B) this automatically selected half of the features: lip, alm1, alm2
select.fit(x, y)
x_selected_bypercent = select.transform(x)
x_chosen1 = select.get_support(
) #this shows the 3 features selected from the 7 possible
print(x_chosen1)
print(x_selected_bypercent
) #the 3 automatically selected features to be used for the train data
#(II)Recursive Feature Elimination - to keep the most important features, important for some Machine learning algorithms.
print(
df.corr(method='pearson')
) #other options: 'kendall', 'spearman'. highest correlation in descending order: mcg, gvh, alm1, lip, aac, chg, alm2.
estimator = SVR(
kernel="linear"
) #Choose the model it is appropriate for: ‘linear’, ‘poly’, ‘rbf’, ‘sigmoid’, ‘precomputed’.
select = RFE(estimator, n_features_to_select=3,
features_train = vectorizer.fit_transform(features_train)
features_test = vectorizer.transform(features_test)
feature_names = vectorizer.get_feature_names()
print(features_train.shape)
print(features_test.shape)
#Feature selection
selector = SelectPercentile(f_classif, percentile=10)
selector.fit(features_train, labels_train)
features_train_transformed = selector.transform(features_train)
features_test_transformed = selector.transform(features_test)
features_train_transformed.shape
print("No of features after selection :", features_train_transformed.shape[1])
#Using MultinomialNB
clf = MultinomialNB()
grid_param = {'alpha': [0.001, 0.01, 0.1, 0.5, 1, 10]}
grid_search = GridSearchCV(estimator=clf,
param_grid=grid_param,
cv=5,
optimum_complexity = complexities[np.argmax(uar_scores)]
print('\nOptimum complexity: {0:.6f}, maximum UAR on Devel {1:.1f}\n'.format(
optimum_complexity,
np.max(uar_scores) * 100))
if not feat_selection:
clf = svm.LinearSVC(C=optimum_complexity, random_state=0)
clf.fit(X_train, y_train)
y_pred = clf.predict(X_test)
else:
uar = []
percentile = [10, 20, 30, 40, 50, 60, 70, 80, 90, 100]
for p in percentile:
selection = SelectPercentile(f_classif, percentile=p)
feat_selected = selection.fit_transform(X_train, y_train)
feat_devel = selection.transform(X_devel)
print('\nComplexity {0:.6f}'.format(optimum_complexity))
#clf = svm.LinearSVC(C=optimum_complexity, random_state=0)
clf = svm.SVC(C=optimum_complexity, kernel='linear', random_state=0)
clf.fit(feat_selected, y_train)
y_pred = clf.predict(feat_devel)
uar.append(
recall_score(y_devel, y_pred, labels=classes, average='macro'))
print('UAR on Devel {0:.1f}'.format(uar[-1] * 100))
if show_confusion:
print('Confusion matrix (Devel):')
print(classes)
print(confusion_matrix(y_devel, y_pred, labels=classes))
optimum_percentile = percentile[np.argmax(uar)]
print(
'\nOptimum percentile: {0:.6f}, maximum UAR on Devel {1:.1f}\n'.format(
def col_filter(mtx_train, y_train, mtx_test, func = chi2, percentile = 90):
feature_select = SelectPercentile(func, percentile = percentile)
feature_select.fit(mtx_train, y_train)
mtx_train = feature_select.transform(mtx_train)
mtx_test = feature_select.transform(mtx_test)
return mtx_train, mtx_test
"_train.npy")
X_test = load_numpy_matrix(feature_set_path +
"Google_TfidfFeatures" + tag +
"_test.npy")
print "\nFeatures", FEATURES[featureV]
print '\nTotal:', X_train.shape[0] + X_test.shape[0]
print 'Features:', X_train.shape[1]
print "\nClass distribution", Counter(y_train)
print "\nClass distribution", Counter(y_test)
# FEATURE SELECT
if featureV == 0:
selector = SelectPercentile(score_func=f_classif,
percentile=perc).fit(X_train, y_train)
X_train = selector.transform(X_train)
X_test = selector.transform(X_test)
elif featureV < 2:
selector = SelectKBest(
score_func=chi2,
k=min(200000, int(X_train.shape[1] * (perc / 100.0)))).fit(
X_train, y_train)
X_train = selector.transform(X_train)
X_test = selector.transform(X_test)
print X_train.shape
print X_test.shape
# FEATURE SCALING
if featureV == 0:
scaler = preprocessing.StandardScaler().fit(X_train)
'csr', 'bool')
print('cv prepared !')
sparse.save_npz(path + '/feature/base_train_csr.npz', base_train_csr)
sparse.save_npz(path + '/feature/base_predict_csr.npz', base_predict_csr)
train_csr = sparse.hstack(
(sparse.csr_matrix(train_x[num_feature]), base_train_csr),
'csr').astype('float32')
predict_csr = sparse.hstack(
(sparse.csr_matrix(predict_x[num_feature]), base_predict_csr),
'csr').astype('float32')
print(train_csr.shape)
feature_select = SelectPercentile(chi2, percentile=50)
feature_select.fit(train_csr, train_y)
train_csr = feature_select.transform(train_csr)
predict_csr = feature_select.transform(predict_csr)
print('feature select')
print(train_csr.shape)
n = 1500
data_col = pd.read_csv('col_sort_one11.csv', header=None)
col = data_col[0].values.copy()
lgb_model = lgb.LGBMClassifier(boosting_type='gbdt',
num_leaves=60,
max_depth=-1,
learning_rate=0.1,
n_estimators=n,
max_bin=425,
subsample_for_bin=50000,
def cross(train_path, test_path, select_feat):
print('Read train and test')
train = pd.read_csv("./data/train_no_events.csv",
dtype={'device_id': np.str})
train.drop(['age', 'gender'], axis=1, inplace=True)
train_label = train["group"]
lable_group = LabelEncoder()
train_label = lable_group.fit_transform(train_label)
test = pd.read_csv("./data/test_no_events.csv",
dtype={'device_id': np.str})
test["group"] = np.nan
trf = open(train_path, 'rb')
train_sp = pickle.load(trf)
trf.close()
ttf = open(test_path, 'rb')
test_sp = pickle.load(ttf)
ttf.close()
train_sp = train_sp.toarray()
if select_feat == "1":
X_train, X_val, y_train, y_val = train_test_split(train_sp,
train_label,
train_size=.90,
random_state=10)
print X_train.shape
print train.shape
print X_val.shape
print("# Feature Selection")
selector = SelectPercentile(f_classif, percentile=100)
selector.fit(X_train, y_train)
X_train = selector.transform(X_train)
X_val = selector.transform(X_val)
print X_train.shape
print X_val.shape
train_sp = selector.transform(train_sp)
test_sp = selector.transform(test_sp)
dtrain = xgb.DMatrix(X_train, y_train)
dvalid = xgb.DMatrix(X_val, y_val)
#dtrain = xgb.DMatrix(train_sp, train_label)
params = {
"objective": "multi:softprob",
"num_class": 12,
"booster": "gblinear",
"eval_metric": "mlogloss",
"eta": 0.05,
"silent": 1,
"lambda": 3,
"alpha": 2,
}
params2 = {
"objective": "multi:softprob",
"num_class": 12,
"booster": "gbtree",
"eval_metric": "mlogloss",
"eta": 0.05,
"max_depth": 6,
"subsample": 0.7,
"colsample_bytree": 0.7,
"num_parallel_tree": 1,
"seed": 114,
"silent": 1,
}
watchlist = [(dtrain, 'train'), (dvalid, 'eval')]
gbm = xgb.train(params2,
dtrain,
1000,
evals=watchlist,
early_stopping_rounds=10,
verbose_eval=True)
else:
selector = SelectPercentile(f_classif, percentile=100)
selector.fit(train_sp, train_label)
#X_train = selector.transform(X_train)
#X_val = selector.transform(X_val)
train_sp = selector.transform(train_sp)
test_sp = selector.transform(test_sp)
#dtrain = xgb.DMatrix(X_train, y_train)
#dvalid = xgb.DMatrix(X_val, y_val)
dtrain = xgb.DMatrix(train_sp, train_label)
dtest = xgb.DMatrix(test_sp)
params = {
"objective": "multi:softprob",
"num_class": 12,
"booster": "gbtree",
"eval_metric": "mlogloss",
"eta": 0.05,
"max_depth": 8,
"subsample": 0.7,
"colsample_bytree": 0.7,
"num_parallel_tree": 1,
"seed": 114,
"silent": 1,
}
params2 = {
"objective": "multi:softprob",
"num_class": 12,
"booster": "gblinear",
"max_depth": 6,
"eval_metric": "mlogloss",
"eta": 0.05,
"silent": 1,
"lambda": 3,
"alpha": 2,
}
res = xgb.cv(params2,
dtrain,
num_boost_round=700,
nfold=5,
callbacks=[
xgb.callback.print_evaluation(show_stdv=False),
xgb.callback.early_stop(3)
])
print(res)
def train_classifier_use_feature_selection(self):
#################
# feature selection
#################
selector = SelectPercentile(score_func=_score_func,
percentile=_percentile)
print("Fitting data with feature selection ...")
selector.fit(x_train, y_train)
# get how many features are left after feature selection
x_features = selector.transform(x_train)
print("Shape of array after feature selection is " +
str(x_features.shape))
clf = SGDClassifier(loss=_loss,
penalty=_penalty,
alpha=_alpha,
n_iter=_n_iter,
random_state=42).fit(x_features, y_train)
# get the features which are selected and write to file
feature_boolean = selector.get_support(indices=False)
##################
# get cross validation score
##################
scores = cross_val_score(clf,
x_features,
y_train,
cv=10,
scoring='f1_weighted')
print("Cross validation score: " + str(scores))
# Get average performance of classifier on training data using 10-fold CV, along with standard deviation
print("Cross validation accuracy: %0.2f (+/- %0.2f)" %
(scores.mean(), scores.std() * 2))
####################
# test clf on test data
####################
# apply feature selection on test data too
x_test_selector = selector.transform(x_test)
print("Shape of array for test data after feature selection is " +
str(x_test_selector.shape))
y_predicted = clf.predict(x_test_selector)
# print the mean accuracy on the given test data and labels
print("Classifier score on test data is: %0.2f " %
clf.score(x_test_selector, y_test))
print(metrics.classification_report(y_test, y_predicted))
cm = metrics.confusion_matrix(y_test, y_predicted)
print(cm)
return clf
cancer = load_breast_cancer()
import numpy as np #노이즈 생성해서 추가
rng = np.random.RandomState(42)
noise = rng.normal(size=(len(cancer.data), 50))
X_w_noise = np.hstack([cancer.data, noise])
X_train, X_test, y_train, y_test = train_test_split(X_w_noise,
cancer.target,
random_state=0,
test_size=0.5)
select = SelectPercentile(percentile=50) #특성의 50%만 선택
select.fit(X_train, y_train)
X_train_selected = select.transform(X_train)
print("X_train.shape : {}".format(X_train.shape))
print("X_train_selected.shape : {}".format(X_train_selected.shape))
mask = select.get_support() #선택된거 안된거 표시
print(mask) #대체로 원본데이터가 선택됨
import matplotlib.pyplot as plt
plt.matshow(mask.reshape(1, -1), cmap="gray") #흰색 선택 O, 검은색 선택 x
plt.xlabel("feature num")
#전체와 선택 성능 비교
from sklearn.linear_model import LogisticRegression
X_test_selected = select.transform(X_test)
class LocalRegression(object):
"""implements a scikitlearn model that finds nearest geographic neighbours and computes a regression. Defaults
to a global regression if location data is not available.
methods:
fit( data, response, location_data )
predict( data, location_data)
"""
def __init__(self,
k=200,
feature_selection=False,
regressor=LinearRegression,
verbose=False,
params={},
response_f=identity,
inv_response_f=identity):
self.k = k
self.response_f = response_f
self.inv_response_f = inv_response_f
self.zero_coef_ = np.zeros(10000)
self.verbose = verbose
self.feature_selection = feature_selection
self.selector = SelectPercentile(f_regression, 50)
#if regressor is a list of regressor, we need to initialize all of them
if type(regressor) == list:
self.regressor = [
self.regressor[i](**params[i]) for i in range(len(regressor))
]
else:
self.regressor = [regressor(**params)]
def __str__(self):
return "%dK%sLocalRegression" % (
self.k, self.regressor.__name__.replace(" ", ""))
def fit(self, data, response, location_data):
try:
#incase I pass in a numpy array or pandas df
self.data_ = data.values
except:
self.data_ = data
try:
self.response_ = response.values
except:
self.response_ = response
try:
self.location_data_ = location_data.values
except:
self.location_data_ = location_data
self.gnn = geoNN.GeoNNFinder(self.location_data_)
return self
def predict(self, data, location_data, weights=None):
if location_data.shape[0] != data.shape[0]:
raise Exception(
"length of first argument does not equal length of second argument."
)
n = location_data.shape[0]
try:
data = data.values
except:
pass
#reg = self.regressor(**self.params)
prediction = np.zeros(n)
try:
location_data = location_data.values
except:
pass
#weights = weights.values
#argweights_sorted = np.argsort( weights )
for i in range(n):
if (self.verbose and i % 100 == 0):
print i
#how many estimators should we make, proptional to the percentile of the weight.
#naive scheme:
#n_estimators = self.trivial_n_estimators( i, argweights_sorted)
location = location_data[i, :]
#sub_predictions = np.zeros( n_estimators)
#for n_est in range(n_estimators):
if np.any(pandas.isnull(location)):
prediction[i] = self.response_.mean()
else:
inx = self.gnn.find(location[0], location[1], self.k)
sub_data = self.data_[inx, :]
sub_response = self.response_f(self.response_[inx, :])
to_predict = data[i, :]
if self.feature_selection:
sub_data = self.selector.fit_transform(
self.data_[inx, :], sub_response)
to_predict = self.selector.transform(data[i, :])
for reg in self.regressor:
reg.fit(sub_data, sub_response)
try:
self.zero_coef_[np.nonzero(
abs(reg.coef_) < .000001)[0]] += 1
except:
pass
#take the average
prediction[i] = np.array([
reg.predict(to_predict) for reg in self.regressor
]).mean()
#prediction[i] = sub_predictions.mean()
#make sure everything is inside [0-100]
prediction = self.inv_response_f(prediction)
prediction[prediction > 100] = 99
prediction[prediction < 0] = 4
return prediction
def naive_n_estimators(self, i, argweights_sorted):
"""returns the deci-percentile plus 1, i.e. if the weight is the 86th percentile, return 9"""
n = int(float(argweights_sorted[i]) / len(argweights_sorted) * 10) + 1
return n
def trivial_n_estimators(self, i, argweights_sorted):
return 1
print(data_frame)
numpy.seterr(divide='ignore', invalid='ignore')
# Train/Test Split
x_train, x_test, y_train, y_test = train_test_split(data_frame,
target,
random_state=0)
print(f'\nTrain data shape: {x_train.shape}')
print(f'Test data shape{x_test.shape}')
print(f'Target shape {y_test.shape}')
# Feature Selection
selection = SelectPercentile(percentile=50)
selection.fit(x_train, y_train)
x_train_compressed = selection.transform(x_train)
print(f'\nTrain shape after selection: {x_train_compressed.shape}')
selection_status = list(selection.get_support())
print(f'Selection Status: {selection_status} Length: {len(selection_status)}')
x_test_compressed = selection.transform(x_test)
# Printing Selected Column Names
i = 0
selected_columns = []
for status in selection_status:
if status:
selected_columns.append(data_column_names[i])
i += 1
print(f'Selected Columns: {selected_columns} Length: {len(selected_columns)}')
# Applying Linear Regression
# In[40]:
#Accuracy over test set after training for all the features
accuracy = clf.score(test.toarray(), np.asarray(y_test.flatten(), dtype=np.int64))
print("Accuracy of test set: ", accuracy)
# In[56]:
#Selecting the top-p percentile features
p = 0.1
select = SelectPercentile(f_classif, percentile=p)
select.fit(train, y_train)
train_select = select.transform(train)
# In[59]:
#Batch wise training of top-p percentile features
n = len(corpus)
batch = 1000
clf2 = GaussianNB()
test_select = select.transform(test)
start = time.time()
for i in range(int(n/batch)):
s = i*batch
e = (i+1)*batch
label=r'Univariate score ($-Log(p_{value})$)', color='darkorange',
edgecolor='black')
# #############################################################################
# Compare to the weights of an SVM
clf = svm.SVC(kernel='linear')
clf.fit(X, y)
svm_weights = (clf.coef_ ** 2).sum(axis=0)
svm_weights /= svm_weights.max()
plt.bar(X_indices - .25, svm_weights, width=.2, label='SVM weight',
color='navy', edgecolor='black')
clf_selected = svm.SVC(kernel='linear')
clf_selected.fit(selector.transform(X), y)
svm_weights_selected = (clf_selected.coef_ ** 2).sum(axis=0)
svm_weights_selected /= svm_weights_selected.max()
plt.bar(X_indices[selector.get_support()] - .05, svm_weights_selected,
width=.2, label='SVM weights after selection', color='c',
edgecolor='black')
plt.title("Comparing feature selection")
plt.xlabel('Feature number')
plt.yticks(())
plt.axis('tight')
plt.legend(loc='upper right')
plt.show()
from sklearn.feature_selection import SelectPercentile, chi2
from sklearn.model_selection import cross_val_score
import pylab as pl
# 利用 5折CV法 在训练集上对合适的特征选择量进行验证
percentiles = range(1, 100, 2)
results = []
for i in percentiles:
fs = SelectPercentile(score_func=chi2, percentile=i)
x_train_fs = fs.fit_transform(x_train, y_train)
# 由于是5折验证， 所以输出score时是5个
scores = cross_val_score(dt, x_train_fs, y_train, cv=5)
results.append(scores.mean())
results = np.array(list(map(lambda x: round(x, 4), results)))
print(results)
print("the Optimal Number of Features is %d" % (percentiles[results.argmax()]))
pl.plot(percentiles, results)
pl.xlabel("percentile of features")
pl.ylabel("acc")
pl.show()
# 利用得到的最优参数重新训练，并对测试集进行预测
fs = SelectPercentile(score_func=chi2,
percentile=percentiles[results.argmax()])
x_train_fs = fs.fit_transform(x_train, y_train)
selectedFeatures = np.array(vec.feature_names_)[fs.get_support()]
dt.fit(x_train_fs, y_train)
x_test_fs = fs.transform(x_test)
print("the score of DT with filtering features is ",
dt.score(x_test_fs, y_test))
print("Best cross-validation accuracy: {:.2f}".format(grid.best_score_))
print("Test set score: {:.2f}".format(grid.score(X_test, y_test)))
print("Best parameters: {}".format(grid.best_params_))
#. ILLUSTRATING INFORMATION LEAKAGE
import numpy as np
from sklearn.svm import SVC
from sklearn.preprocessing import MinMaxScaler
# Load and split the data
rnd = np.random.RandomState(seed=0)
X = rnd.normal(size=(100, 10000))
y = rnd.normal(size=(100, ))
from sklearn.feature_selection import SelectPercentile, f_regression
select = SelectPercentile(score_func=f_regression, percentile=5).fit(X, y)
X_selected = select.transform(X)
print("X_selected.shape: {}".format(X_selected.shape))
from sklearn.model_selection import cross_val_score
from sklearn.linear_model import Ridge
print("Cross-validation accuracy (cv only on ridge): {:.2f}".format(
np.mean(cross_val_score(Ridge(), X_selected, y, cv=5))))
from sklearn.pipeline import Pipeline
pipe = Pipeline([("select",
SelectPercentile(score_func=f_regression, percentile=5)),
("ridge", Ridge())])
print("Cross-validation accuracy (pipeline): {:.2f}".format(
np.mean(cross_val_score(pipe, X, y, cv=5))))
#CONVENIENT PIPELINE INTERFACE WITH MAKE_PIPELINE
import numpy as np
from sklearn.svm import SVC
import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
from sklearn.feature_selection import SelectPercentile
from sklearn.linear_model import LogisticRegression
cancer = load_breast_cancer()
rng = np.random.RandomState(42)
noise = rng.normal(size=(len(cancer.data),50))
X_w_noise = np.hstack([cancer.data,noise])
X_train, X_test, y_train, y_test = train_test_split(X_w_noise,cancer.target,random_state=0,test_size=0.5)
select = SelectPercentile(percentile=50)
select.fit(X_train,y_train)
X_train_selected = select.transform(X_train)
print("X_train.shape: {}".format(X_train.shape))
print("X_train_selected.shape: {}".format(X_train_selected.shape))
mask = select.get_support()
print(mask)
plt.matshow(mask.reshape(1,-1),cmap="gray_r")
plt.show()
#Initialize and fit scaler
scaler = StandardScaler()
#Fit scaler using the training data
scaler.fit(X_train_raw)
#Transform the raw data
X_train_standardized = scaler.transform(X_train_raw)
X_test_standardized = scaler.transform(X_test_raw)
#Initialize and fit selector
MI_selector = SelectPercentile(mutual_info_classif, percentile=60)#Remove the lower 40%
MI_selector.fit(X_train_standardized, y_train.values.ravel())
#Transform
X_train_MI = MI_selector.transform(X_train_standardized)
X_test_MI = MI_selector.transform(X_test_standardized)
#Summary
print("Feature Selection Results - Univariate Feature Selection")
#Summary
print("Filter Result:")
print("Number of features: ",X_train_MI.shape[1])
#Rank the features by scores
plt.figure(figsize=(10, 8), dpi= 60)
feat_scores = pd.Series(MI_selector.scores_, index=X_train_raw.columns)
top_feat = feat_scores.nlargest(10)
top_feat.plot(kind='barh')
plt.title("Feature Ranking by Mutual Information Score")
