def GaussianNB(feature_list, dataset):
from sklearn.naive_bayes import GaussianNB
clf = GaussianNB()
test_classifier(clf, dataset, feature_list)
#score = clf.
return clf
def RandomForest(feature_list,dataset):
from sklearn.ensemble import RandomForestClassifier
clf = RandomForestClassifier()
test_classifier(clf,dataset,feature_list)
imp= clf.feature_importances_
print_importance (feature_list,imp)
return clf
def tune_classifier(classifier, clf_params, max_features):
### features_list is a list of strings, each of which is a feature name.
### The first feature must be "poi".
features_list = get_feature_list()
### Create new feature(s)
### Store to my_dataset for easy export below.
my_dataset = get_data()
### Extract features and labels from dataset for local testing
features_list = features_list[0:max_features+1]
data, labels, features = get_features_and_labels(my_dataset, features_list)
### Tune your classifier to achieve better than .3 precision and recall
### using our testing script. Check the tester.py script in the final project
### folder for details on the evaluation method, especially the test_classifier
### function. Because of the small size of the dataset, the script uses
### stratified shuffle split cross validation. For more info:
### http://scikit-learn.org/stable/modules/generated/sklearn.cross_validation.StratifiedShuffleSplit.html
from sklearn.cross_validation import train_test_split
features_train, features_test, labels_train, labels_test = \
train_test_split(features, labels, test_size=0.3, random_state=42)
# Testing
clf = GridSearchCV(classifier, param_grid=clf_params, scoring=make_scorer(f1_score))
clf.fit(features_train, labels_train)
clf_final = clf.best_estimator_
print "The best estimator = ", clf_final
test_classifier(clf_final, my_dataset, features_list, 1000)
def decisionTree(feature_list, dataset):
from sklearn import tree
clf = tree.DecisionTreeClassifier()
test_classifier(clf, dataset, feature_list)
print clf.feature_importances_
return clf
def iterPipe(num1, num2):
for i in range(num1, num2 + 1):
# estimators = [('scaling', StandardScaler()),('reduce_dim', PCA()), ('dtc', DTC(min_samples_split=i*2))]
# estimators = [('reduce_dim', PCA(n_components=2)), ('dtc', DTC(min_samples_split=i))]
# clfIter = Pipeline(estimators)
# clfIter.set_params(reduce_dim__n_components=3)
clfIter = DTC(min_samples_split=i)
test_classifier(clfIter, my_dataset, features_list)
def KNN(feature_list,dataset):
from sklearn.neighbors import KNeighborsClassifier
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
knn = KNeighborsClassifier()
# feature scale
estimators = [('scale', StandardScaler()), ('knn', knn)]
clf = Pipeline(estimators)
test_classifier(clf, my_dataset, features_list)
def setup_and_test(my_dataset, features_list, classifier):
# Dump classifier and features list, so we can test them
dump_classifier_and_data(classifier, my_dataset, features_list)
# load up student's classifier, dataset, and feature_list
clf, dataset, feature_list = load_classifier_and_data()
# Run testing script
test_classifier(clf, dataset, feature_list)
return
def tuneDT(feature_list,dataset):
from sklearn.neighbors import KNeighborsClassifier
from sklearn.grid_search import GridSearchCV
from sklearn import tree
tree_clf = tree.DecisionTreeClassifier()
parameters = {'criterion':('gini', 'entropy'),
'splitter':('best','random')}
clf = GridSearchCV(tree_clf, parameters,scoring = 'recall')
test_classifier(clf, my_dataset, features_list)
print '###best_params'
print clf.best_params_
def detect_poi():
### Load the dictionary containing the dataset
data_dict = pickle.load(open("final_project_dataset.pkl", "r") )
### Task 1: Remove outliers
data_dict.pop('TOTAL',0)
### Task 2: Select what features
### 'stk_pay_ratio','to_poi_ratio', 'from_poi_ratio','bonus_salary_ratio'
### features_list is a list of strings, each of which is a feature name.
### The first feature must be "poi".
my_dataset = data_dict
stk_pay_ratio(my_dataset)
from_poi_ratio(my_dataset)
to_poi_ratio(my_dataset)
bonus_salary_ratio(my_dataset)
### Task 3: Feature Selection
### Generate a set of 15 feature lists from these 4 features
### This way, all possible combinations of these features are tested
all_features_list = fList_set()
### Because of the small size of the dataset, the script uses stratified
### shuffle split cross validation in tester.py
metrics = []
clf = GaussianNB()
### ptest uses Stratified shuffle split cross validation and calculates the precision
### Find the precision for every list
for i in range(0,15):
metrics.append(ptest(clf,my_dataset,all_features_list[i]))
### Go for the feature list that produces the best precision.
### For this dataset only, it is harder to get a high precision.
best = np.array(metrics).argmax()
### Run test_classifier to print evaluation metrics to console
test_classifier(clf, my_dataset,all_features_list[best])
### Now use the same feature list to run the decison tree classifier
features_list = all_features_list[best]
### Task 4: Try a varity of classifiers
samples_split_values = [2,4]
samples_leaf_values = [1,2]
for split in samples_split_values:
for leaf in samples_leaf_values:
clf = tree.DecisionTreeClassifier(min_samples_split=split,\
min_samples_leaf=leaf)
test_classifier(clf, my_dataset, features_list)
print_feature_importances(features_list, clf)
###Choose best classfier and feature set
clf = GaussianNB()
### Dump classifier, dataset, and features_list
dump_classifier_and_data(clf, my_dataset, features_list)
def tuneKmeans(feature_list,dataset):
from sklearn.cluster import KMeans
from sklearn.grid_search import GridSearchCV
km_clf = KMeans(n_clusters=2, tol=0.001)
parameters = {'n_clusters': (2,10)}
clf = GridSearchCV(km_clf, parameters, scoring='recall')
test_classifier(clf, dataset, feature_list)
print '###best_params'
print clf.best_params_
return clf.best_estimator_
def explore_scores():
for n in features:
for c in n_neighbor:
for d in weights:
for e in algorithm:
for f in leaf_size:
for g in p:
for h in metric:
feature = 0
feature = features_select(n)
pipeline = Pipeline([('normalization', scaler),
('classifier', KNeighborsClassifier(n_neighbors=c, weights=d, algorithm=e,
leaf_size=f, p=g, metric=h))])
test_classifier(pipeline, enron_data, feature)
def tuneKNN(feature_list,dataset):
from sklearn.neighbors import KNeighborsClassifier
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.grid_search import GridSearchCV
knn = KNeighborsClassifier()
# feature scale
estimators = [('scale', StandardScaler()), ('knn', knn)]
pipeline = Pipeline(estimators)
parameters = {'knn__n_neighbors':[1,8],
'knn__algorithm':('ball_tree','kd_tree','brute','auto')}
clf = GridSearchCV(pipeline, parameters,scoring = 'recall')
test_classifier(clf, my_dataset, features_list)
print '###best_params'
print clf.best_params_
def getRF():
print "==============="
print "RandomForests"
print "==============="
for score in scores:
print score
print
#parameters = {'n_estimators':range(10, 150, 10), 'criterion':['gini', 'entropy'], 'min_samples_split':range(2, 8, 2)}
parameters = {'rf__n_estimators':range(10, 150, 10), 'rf__criterion':['gini', 'entropy'], 'rf__min_samples_split':range(2, 8, 2),
'selector__k':range(3, 22, 1)}
gs = grid_search.GridSearchCV(rf_pipe, parameters, scoring=score, cv=cv)
gs.fit(features, labels)
#This is the model you pass to tester.py
clf = gs.best_estimator_
print " "
print "Optimal Model - by Grid Search"
print clf
print " "
best_parameters = gs.best_estimator_.get_params()
print " "
print "Best Parameters- by Grid Search"
print best_parameters
print " "
labels_pred = gs.predict(features)
# Print Results (will print the Grid Search score)
print "Grid Search Classification report:"
print " "
print classification_report(labels, labels_pred)
print ' '
# Print Results (will print the tester.py score)
print "tester.py Classification report:"
print " "
test_classifier(clf, my_dataset, features_list)
print " "
print
def getAda():
print "==============="
print "AdaBoost"
print "==============="
for score in scores:
print score
print
#parameters = {'n_estimators':range(50, 100, 1), 'learning_rate':[x * 0.01 for x in range(100, 160, 1)]}
parameters = {'ada__n_estimators': range(1, 100, 20), 'ada__learning_rate':[x * 0.01 for x in range(100, 160, 10)],
'selector__k':range(3, 22, 1)}
gs = grid_search.GridSearchCV(ada_pipe, parameters, scoring=score, cv=cv)
gs.fit(features, labels)
#This is the model you pass to tester.py
clf = gs.best_estimator_
print " "
print "Optimal Model - by Grid Search"
print clf
print " "
best_parameters = gs.best_estimator_.get_params()
print " "
print "Best Parameters- by Grid Search"
print best_parameters
print " "
labels_pred = gs.predict(features)
# Print Results (will print the Grid Search score)
print "Grid Search Classification report:"
print " "
print classification_report(labels, labels_pred)
print ' '
# Print Results (will print the tester.py score)
print "tester.py Classification report:"
print " "
test_classifier(clf, my_dataset, features_list)
print " "
print
def getKNN():
print "==============="
print "KNeighborsClassifier"
print "==============="
for score in scores:
print score
print
#parameters = {'n_neighbors':range(2, 10, 2), 'weights':['distance', 'uniform'], 'metric':['minkowski', 'euclidean']}
parameters = {'knn__n_neighbors': range(2, 10, 2), 'knn__weights':['distance', 'uniform'], 'knn__metric':['minkowski', 'euclidean'],
'selector__k':range(3, 20, 1)}
gs = grid_search.GridSearchCV(knn_pipe, parameters, scoring=score, cv=cv)
gs.fit(features, labels)
#This is the model you pass to tester.py
clf = gs.best_estimator_
print " "
print "Optimal Model - by Grid Search"
print clf
print " "
best_parameters = gs.best_estimator_.get_params()
print " "
print "Best Parameters- by Grid Search"
print best_parameters
print " "
labels_pred = gs.predict(features)
# Print Results (will print the Grid Search score)
print "Grid Search Classification report:"
print " "
print classification_report(labels, labels_pred)
print ' '
# Print Results (will print the tester.py score)
print "tester.py Classification report:"
print " "
test_classifier(clf, my_dataset, features_list)
print " "
print
def getSVC():
print "==============="
print "SVC"
print "==============="
for score in scores:
print score
print
parameters = {'sv__C': [0.01, 0.1, 1, 500, 1000, 5000, 10000, 50000, 100000], 'sv__kernel':['linear'],
'selector__k':range(3, 22, 1)} #'sv__gamma':[0.0001, 0.0005, 0.001, 0.005, 0.01, 0.1, 1, 10, 100, 500, 1000],
gs = grid_search.GridSearchCV(sv_pipe, parameters, scoring=score, cv=cv)
gs.fit(features, labels)
#This is the model you pass to tester.py
clf = gs.best_estimator_
print " "
print "Optimal Model - by Grid Search"
print clf
print " "
best_parameters = gs.best_estimator_.get_params()
print " "
print "Best Parameters- by Grid Search"
print best_parameters
print " "
labels_pred = gs.predict(features)
# Print Results (will print the Grid Search score)
print "Grid Search Classification report:"
print " "
print classification_report(labels, labels_pred)
print ' '
# Print Results (will print the tester.py score)
print "tester.py Classification report:"
print " "
test_classifier(clf, my_dataset, features_list)
print " "
print
def getNB():
print "==============="
print "GaussianNB"
print "==============="
for score in scores:
print score
print
parameters = {'selector__k':range(3, 22, 1)}
gs = grid_search.GridSearchCV(nb_pipe, parameters, scoring=score, cv=cv)
gs.fit(features, labels)
#This is the model you pass to tester.py
clf = gs.best_estimator_
print " "
print "Optimal Model - by Grid Search"
print clf
print " "
best_parameters = gs.best_estimator_.get_params()
print " "
print "Best Parameters- by Grid Search"
print best_parameters
print " "
labels_pred = gs.predict(features)
# Print Results (will print the Grid Search score)
print "Grid Search Classification report:"
print " "
print classification_report(labels, labels_pred)
print ' '
# Print Results (will print the tester.py score)
print "tester.py Classification report:"
print " "
test_classifier(clf, my_dataset, features_list)
print " "
print
def train_test():
data = featureFormat(my_dataset, features_list, sort_keys = True)
labels, features = targetFeatureSplit(data)
features_train, features_test, labels_train, labels_test = train_test_split(features, labels, test_size=0.3, random_state=42)
clf = DecisionTreeClassifier(random_state=42)
clf.fit(features_train, labels_train)
print test_classifier(clf, my_dataset, features_list)
### Print feature importance in order
features_imp = {}
for i in xrange(len(features_list)-1):
features_imp[features_list[1+i]] = clf.feature_importances_[i]
pprint(sorted(features_imp.items(), key=operator.itemgetter(1),reverse=True))
def main():
data_dict = pickle.load(open("final_project_dataset.pkl", "r"))
my_dataset = data_dict
my_dataset = AddFeatures(my_dataset)
# Exclude using Discretion.
Exc1 = ["email_address"]
# Replaced by creating better versions of the features
Exc2 = ["to_messages", "from_messages", "from_this_person_to_poi", "from_poi_to_this_person"]
# Exclude because Highly Correlated with stronger features
Exc3 = [
"deferral_payments",
"expenses",
"deferred_income",
"restricted_stock_deferred",
"director_fees",
"long_term_incentive",
"bonus",
"total_payments",
"salary",
"total_stock_value",
"restricted_stock",
"exercised_stock_options",
"other",
]
exclude = Exc1 + Exc2 + Exc3
# QueryDataSet(my_dataset)
# ShowCorrel(my_dataset)
features_list = next(my_dataset.itervalues()).keys()
for i in exclude:
features_list.remove(i)
features_list.insert(0, features_list.pop(features_list.index("poi")))
data = featureFormat(my_dataset, features_list, sort_keys=True)
### Extract features and labels from dataset for local testing
labels, features = targetFeatureSplit(data)
features_train, features_test, labels_train, labels_test = train_test_split(
features, labels, test_size=0.1, random_state=42, stratify=labels
)
# clf=TuneSVM(features, labels,features_list)
# clf=TuneKNN(features, labels,features_list)
# clf=NoTuneDT(features, labels,features_list)
# clf=TuneDT(features,labels,features_list)
features_list.insert(0, "poi")
dump_classifier_and_data(clf, my_dataset, features_list)
test_classifier(clf, my_dataset, features_list)
def train_and_predict(first,second):
#trains the model and returns the value of desired evaluation metric
features_list = ["poi",first,second]
data = featureFormat(my_dataset, features_list, sort_keys = True)
labels, features = targetFeatureSplit(data)
from sklearn.naive_bayes import GaussianNB
from sklearn import tree
if dt:
clf = tree.DecisionTreeClassifier()
else:
clf = GaussianNB()
if f1:
return test_classifier(clf, my_dataset, features_list,return_F1=True)
else:
return test_classifier(clf, my_dataset, features_list,return_precision=True)
def get_top_features_all_data(X_df, y_df, grid_searcher, top_N=9):
'''Give an estimate of the model produced by grid_search using features
selected from using ExtraTreesClassifier on the entire dataset before
searching for a model.
In general, this may produce overly optimistic results since there is
leakage from the test dataset when selecting features using the entire
dataset.
This is to show that this can improve cross-validated internal testing
over choosing kbest within each cross-validation fold, but is still
overly optimistic if the model were to be used on completely new data.
Args:
X_df: Pandas dataframe of features used to predict.
y_df: Pandas dataframe of labels being predicted.
grid_searcher: GridSearchCV object being searched over for optimal
tuning parameters.
top_N: Top N features to retain based on feature importances obtained
from the ExtraTreesClassifier estimator used in the
top_N_features() function.
Returns:
A list of the top N features that were selected to be fed into the
GridSearchCV object.
Prints:
Test results from the 1000 cross-validation splits testing in tester.py
'''
top_N_features = top_importances(X_df, y_df, top_N=top_N)
top_N_names = list(top_N_features.index)
X_df = X_df[top_N_names]
features_list = ['poi'] + list(top_N_names)
grid_searcher.fit(X_df, y_df)
clf = grid_searcher.best_estimator_
my_dataset = combine_to_dict(features_df=X_df, labels_df=y_df)
data = featureFormat(my_dataset, features_list, sort_keys = True)
labels, features = targetFeatureSplit(data)
test_classifier(clf, my_dataset, features_list)
return top_N_features
def find_best_features(feature_names, features, labels, classifier_fun, search_grid, normalize_data=False):
results = []
processed_features = np.array(features)
processed_labels = labels
if normalize_data:
scaler = StandardScaler()
processed_features = scaler.fit_transform(processed_features, processed_labels)
feature_selector = SelectKBest(k="all")
feature_selector.fit(processed_features, processed_labels)
ranked_features = sorted(zip(feature_names, feature_selector.scores_), key=lambda t: t[1], reverse=True)
ranked_feature_names = [t[0] for t in ranked_features]
logging.info("Scored features:\n%s", pprint.pformat(ranked_features))
logging.info("Ranked feature names: %s", ranked_feature_names)
for k in range(1, len(feature_names) + 1):
logging.info("Selecting %s best feature(s)", k)
selected_feature_names = ranked_feature_names[:k]
logging.info("Selected features: %s", selected_feature_names)
feature_indices = [feature_names.index(f) for f in selected_feature_names]
feature_subset = processed_features[:, feature_indices]
clf = classifier_fun(random_state=98123)
logging.info("Tuning classifier parameters.")
clf_tune = grid_search.GridSearchCV(
clf, search_grid, n_jobs=-1, cv=StratifiedShuffleSplit(labels, n_iter=1000, random_state=42), scoring="f1"
)
clf_tune.fit(feature_subset, processed_labels)
logging.info("Scores:\n%s", pprint.pformat(clf_tune.grid_scores_))
logging.info("Best parameters: %s with score %s", clf_tune.best_params_, clf_tune.best_score_)
clf = classifier_fun(random_state=1987341, **clf_tune.best_params_)
logging.info("Testing classifier.")
precision, recall, f1 = test_classifier(clf, feature_subset, processed_labels)
results.append((k, precision, recall, f1))
logging.info("Best features:\n%s", pprint.pformat(results))
def analyze_feats(each_feature_set, my_dataset, scoresheet_highest_accuracy,
scoresheet_highest_precision):
data = featureFormat(my_dataset, each_feature_set, sort_keys=True)
labels, features = targetFeatureSplit(data)
features_train, features_test, labels_train, labels_test = (
train_test_split(features, labels, test_size=0.5, random_state=42))
# ################## For each feature set, tune the SVC parameter and
# return the best SVC parameters
# tuned_parameters = [{'kernel': ['rbf'], 'C': [1, 3, 10, 100, 1000],
# 'degree':[1,2,3]}]
# #score = 'precision'
# clf = GridSearchCV(SVM, tuned_parameters)
# clf.fit(features_train, labels_train)
# SVM = clf.best_estimator_
# print SVM
# For each feature set, tune the SVC parameter and return the best SVC
# parameters
DT_tuned_parameters = [{'min_samples_split': [30, 40, 50]}]
# score = 'precision'
dt_clf = GridSearchCV(tree.DecisionTreeClassifier(), DT_tuned_parameters)
dt_clf.fit(features_train, labels_train)
DT = dt_clf.best_estimator_
print DT
classifier_type = [DT]
# continue
# run each type of classifier and return results
try:
total_results = []
for index, each_clf in enumerate(classifier_type):
results = test_classifier(each_clf, my_dataset, each_feature_set)
print each_feature_set, results
total_results.append(results)
# for a given feature set, find the classifier with highest
# precision/accuracy and store it in a list
for index, num in enumerate(total_results):
if num[1] == max([accuracy[1] for accuracy in total_results]):
# print "Highest accuracy: \t", num[0], num[1]
scoresheet_highest_accuracy.append(
[each_feature_set, total_results[index]])
if num[1] == max([precision[1] for precision in total_results]):
# print "Highest precision: \t", num[0], num[1]
scoresheet_highest_precision.append(
[each_feature_set, total_results[index]])
except:
pass
def try_all_k_best(max=13):
data = featureFormat(my_dataset, features_list, sort_keys = True)
labels, features = targetFeatureSplit(data)
features_train, features_test, labels_train, labels_test = \
train_test_split(features, labels, test_size=0.25, random_state=42)
for k in range(1,max+1):
pipe = Pipeline([('impute', Imputer(strategy='median')),
('select', SelectKBest(k=k)),
('classify', LogisticRegressionCV())])
pipe.fit(features_train, labels_train)
total_predictions, accuracy, precision, recall, f1, f2 = \
test_classifier(pipe, my_dataset, features_list, folds=1000)
acc.append(accuracy)
prec.append(precision)
reca.append(recall)
def one_feature_predict(features_list, my_dataset):
all = []
for i in features_list:
if i != 'poi':
l = []
l.append('poi')
l.append(i)
all.append(l)
#print all
mycolumns = ['feature_list', 'accuracy', 'precision', 'recall', 'f1', 'f2']
resultdf = pd.DataFrame(columns=mycolumns)
for item in all:
data = featureFormat(my_dataset, item, sort_keys = True)
labels, features = targetFeatureSplit(data)
clf = tree.DecisionTreeClassifier(min_samples_split = 4)
clf.fit(features, labels)
resultdf.loc[len(resultdf)] = (test_classifier(clf, my_dataset, item))
return resultdf
def analyze_feats(each_feature_set, my_dataset, classifier_type, scoresheet_highest_accuracy,
scoresheet_highest_precision, scoresheet_highest_recall):
# run each type of classifier and return results
try:
total_results = []
for index, each_clf in enumerate(classifier_type):
results, feature_importances = test_classifier(each_clf, my_dataset, each_feature_set)
if len(feature_importances) > 0:
# results, feature_importances = test_classifier(each_clf, my_dataset, each_feature_set)
print "####CLF NAME",each_clf
print "#####Length of feature_importances",len(feature_importances)
np.asarray(feature_importances)
importances = zip(np.mean(feature_importances, axis=0),each_feature_set[1:])
importances = sorted(importances,key=lambda i:i[0],reverse=True)
print "#####Length of importances",len(importances)
print importances
print each_feature_set, results
total_results.append(results)
print "total_results",total_results
# for a given feature set, find the classifier with highest
# precision/accuracy and store it in a list
for index, num in enumerate(total_results):
if num[1] == max([accuracy[1] for accuracy in total_results]):
print "Highest accuracy: \t", num[0], num[1]
scoresheet_highest_accuracy.append(
[each_feature_set, total_results[index]])
if num[2] == max([precision[2] for precision in total_results]):
print "Highest precision: \t", num[0], num[2]
scoresheet_highest_precision.append(
[each_feature_set, total_results[index]])
if num[3] == max([recall[3] for recall in total_results]):
print "Highest recall: \t", num[0], num[3]
scoresheet_highest_recall.append(
[each_feature_set, total_results[index]])
except:
pass
return
def tuneNB():
for i in range(1, 20):
acc = []
prec = []
reca = []
testing_features_list = [u'poi']
for feature in features_list_score_order:
testing_features_list.append(feature)
pipe = Pipeline([('impute', Imputer(strategy='median')),
('classify', GaussianNB(priors=[(i/2.)*.1, (1 - (i/2.)*.1)]))])
total_predictions, accuracy, precision, recall, f1, f2 = \
test_classifier(pipe, my_dataset, testing_features_list, folds=200)
acc.append(accuracy)
prec.append(precision)
reca.append(recall)
acc_all.append(acc)
prec_all.append(prec)
reca_all.append(reca)
results_dict['prec' + str(i)] = prec
results_dict['reca' + str(i)] = reca
results_dict['acc' + str(i)] = acc
'min_samples_leaf': [x for x in range(1, 10, 2)],
'max_depth': [None, 1, 2, 4, 8, 12, 18],
'max_features': ['log2', 'sqrt']
}
rfc = RandomForestClassifier(random_state=42)
clf = GridSearchCV(rfc, param_list, cv=5, verbose=3, n_jobs=-1)
clf_ = clf.fit(features, labels)
print clf.best_score_
print clf.best_estimator_
print "Training Set Score:", clf.score(X_train, y_train)
print "Validation Set Score:", clf.score(X_val, y_val)
## Cross Validation of Model
test_classifier(clf.best_estimator_, features, labels, folds=100)
###########################################################################
## Make predictions
test_features = scl.fit_transform(
df_test.drop(['Survived', 'PassengerId'], axis=1).values)
clf.best_estimator_.fit(features, labels)
predictions = clf.best_estimator_.predict(test_features)
output_df = pd.DataFrame({
'PassengerId': df_test['PassengerId'],
'Survived': pd.Series(predictions)
})
output_df = output_df.astype('Int64')
'total_stock_value', 'prop_to_poi', 'prop_from_poi' ] # You will need to use more features from tester import test_classifier from sklearn.ensemble import RandomForestClassifier from sklearn.linear_model import SGDClassifier from sklearn.naive_bayes import GaussianNB from sklearn.tree import DecisionTreeClassifier clf = DecisionTreeClassifier(random_state=42) clf2 = GaussianNB() clf3 = RandomForestClassifier(random_state=42) clf4 = SGDClassifier(random_state=42) test_classifier(clf, fp, features_list, folds=1000) test_classifier(clf2, fp, features_list, folds=1000) test_classifier(clf3, fp, features_list, folds=1000) test_classifier(clf4, fp, features_list, folds=1000) ### Result from initial classifier using all features ### DecisionTreeClassifier Accuracy: 0.80840 Precision: 0.26327 Recall: 0.24300 F1: 0.25273 F2: 0.24680 #BEST higher overall precision, recall& F1 ### GaussianNB Accuracy: 0.83920 Precision: 0.32890 Recall: 0.19800 F1: 0.24719 F2: 0.21512 ### RandomForestClassifier Accuracy: 0.86073 Precision: 0.42811 Recall: 0.13250 F1: 0.20237 F2: 0.15373 ### SGDClassifier Accuracy: 0.52980 Precision: 0.10665 Recall: 0.34250 F1: 0.16265 F2: 0.23747 # Using Decision Tree Classifier to find attributes of importance fp = pd.DataFrame(fp) fp = fp.transpose() X = fp.drop(['poi'], axis=1)
#dt = DecisionTreeClassifier()
t0 = time()
grid_obj = GridSearchCV(dt, parameters, scoring='f1', cv=sss)
print "======== Decision Tree (Optimized) ========"
print("DecisionTree tuning: %r" % round(time() - t0, 3))
# TODO: Fit the grid search object to the training data and find the optimal parameters
t0 = time()
grid_obj = grid_obj.fit(features, labels)
print("DecisionTree fitting: %r" % round(time() - t0, 3))
# Get the estimator
dt = grid_obj.best_estimator_
## Print the parameters
print dt.get_params(), '\n'
print 'Result of feature_list without new create feature:'
test_classifier(dt,
my_dataset,
features_list_without_create_feature,
folds=100)
print 'Result of feature_list with new create feature:'
test_classifier(dt, my_dataset, features_list, folds=100)
clf = dt
### Task 6: Dump your classifier, dataset, and features_list so anyone can
### check your results. You do not need to change anything below, but make sure
### that the version of poi_id.py that you submit can be run on its own and
### generates the necessary .pkl files for validating your results.
dump_classifier_and_data(clf, my_dataset, features_list)
clf = clf.fit(feature_train, target_train)
accuracy_grid = clf.score(feature_train, target_train)
print "Best estimator found by grid search:"
print clf.best_estimator_
return clf.best_estimator_, accuracy_grid
###############################################################################
enron_method_svm = SVR(kernel='rbf')
param_grid_svm = {
'C': [1e3, 5e3, 1e4, 5e4, 1e5],
'gamma': [0.0001, 0.0005, 0.001, 0.005, 0.01, 0.1],
}
[clf_grid, acc5] = val_grid(enron_method_svm, features_train, labels_train,
param_grid_svm)
print acc5
[accuracy, precision, recall, f1,
f2] = test_classifier(clf_svm,
pd.DataFrame(data_dict),
features_list,
folds=1000)
### Task 6: Dump your classifier, dataset, and features_list so anyone can
### check your results. You do not need to change anything below, but make sure
### that the version of poi_id.py that you submit can be run on its own and
### generates the necessary .pkl files for validating your results.
dump_classifier_and_data(clf_reg, my_dataset, features_list)
# parameters = {
# 'anova__k': (2,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21),
# 'anova__k': [2,4,6,8,10,12,14,16,18,20],
# 'pca__whiten': [True, False],
# 'pca__n_components': [6,8,10,11], # For use with PCA
# 'clf__min_samples_split':[2,10,20,30,40,50], # for use with DecisionTree
# 'clf__criterion': ['gini','entropy'] # for use with DecisionTree
# 'clf__n_estimators': [50,100,200], # For use with Adaboost
# 'clf__C': [1, 10, 100, 1e3, 5e3, 1e4, 5e4, 1e5], # for use with SVM
# 'clf__gamma': [0.0001, 0.0005, 0.001, 0.005, # for use with SVM
# 0.01, 0.1],
# 'clf__kernel': ['linear','rbf','poly'] # for use with SVM
# }
## Create Cross Validation object for use in GridSearchCV
# cv = StratifiedShuffleSplit(labels, 1000, random_state = 42)
## Apply GridSearchCV to the dataset
# clf = GridSearchCV(clf, parameters, scoring = 'f1', cv=cv)
# clf.fit(features, labels)
## Set the best performing combination of parameters as the new classifier
# clf = clf.best_estimator_
## Use included tester function to assess performance using cross validation
test_classifier(clf, my_dataset, features_list)
### Dump your classifier, dataset, and features_list so
### anyone can run/check your results.
dump_classifier_and_data(clf, my_dataset, features_list)
## The DT algorithm results will be displayed while running tester.py
# dt_pred = dt_clf.predict(features_test)
# print "DT best accuracy:", accuracy_score(dt_pred,labels_test)
# print "DT Precision:", precision_score(labels_test, dt_pred)
# print "DT Recall:", recall_score(labels_test, dt_pred)
###########################################################################################
## Run tester.py
selected_features = [
'poi', 'total_payments', 'total_stock_value', 'salary', 'bonus',
'fraction_from_poi', 'fraction_to_poi'
]
dump_classifier_and_data(dt_clf_best, enron_less_outliers, selected_features)
test_classifier(dt_clf_best, enron_less_outliers, selected_features)
##########################################################################################
###########################################################################################
# ## Decision Tree Using SelectK in GridSearchCV
#
# print "Check performance of Decision Tree Using SelectK in GridSearchCV"
#
# data = featureFormat(enron_less_outliers, features_all, sort_keys=True)
# labels, features = targetFeatureSplit(data)
#
# features_train, features_test, labels_train, labels_test = cross_validation.train_test_split(
# features, labels, test_size=0.2, random_state=42)
# dt = tree.DecisionTreeClassifier(random_state=42)
#
### folder for details on the evaluation method, especially the test_classifier
### function. Because of the small size of the dataset, the script uses
### stratified shuffle split cross validation. For more info:
### http://scikit-learn.org/stable/modules/generated/sklearn.cross_validation.StratifiedShuffleSplit.html
print "Tuning parameters of classifiers"
print
# Naive Bayes classifier
print "Performing Grid Search of Naive Bayes classification"
param_dict_NB = {'feature_selection__k': range(5, len(features_list))}
gs = grid_search(steps_NB, param_dict_NB, features, labels)
gs_clf_NB = gs.best_estimator_
print '\n Score Metrics Decission Tree Classifier'
test_classifier(gs_clf_NB, data_dict, features_list, folds=1000)
print
# The rest of the parameter tuning is comented out due to time execution
# NB turned out to be the classifier with better scores
'''
# Decision Tree classifier
print "Performing Grid Search of Decission Tree classification"
param_dict_DT = {'feature_selection__k': range(5, len(features_list)),\
'Decission_Tree__criterion': ['gini', 'entropy'],\
'Decission_Tree__min_samples_split' : [2, 3, 4, 5, 6, 7, 8, 9, 10]}
gs = grid_search(steps_DT, param_dict_DT, features, labels)
gs_clf_DT = gs.best_estimator_
print '\n Score Metrics Decission Tree Classifier'
test_classifier(gs_clf_DT, data_dict, features_list, folds = 100)
false_positives, false_negatives,
true_negatives)
print ""
except:
print "Got a divide by zero when trying out:", clf
##running through different fold inputs of k-fold cross-validation
#folds = [2,3,5,10]
#for each in folds:
# test_classifier_kfold(clf,my_dataset,features_list,each)
##test the algorithm multiple times and obtain the accuracy, precision, and recall averages
tot_accuracy = 0
tot_precision = 0
tot_recall = 0
i = 0
while i < 10:
accuracy, precision, recall = test_classifier(clf, my_dataset,
features_list)
tot_accuracy += accuracy
tot_precision += precision
tot_recall += recall
i += 1
print tot_accuracy / float(10), tot_precision / float(10), tot_recall / float(
10)
### Dump your classifier, dataset, and features_list so
### anyone can run/check your results.
dump_classifier_and_data(clf, my_dataset, features_list)
cv = StratifiedShuffleSplit(labels, folds, random_state=42)
for train_idx, test_idx in cv:
features_train = []
features_test = []
labels_train = []
labels_test = []
for ii in train_idx:
features_train.append(features[ii])
labels_train.append(labels[ii])
for jj in test_idx:
features_test.append(features[jj])
labels_test.append(labels[jj])
## Initial algorithms scores
clf_AB = AdaBoostClassifier()
tester.test_classifier(clf_AB, data_dict, features_list)
clf_RBF = SVC(kernel='rbf', max_iter=1000)
tester.test_classifier(clf_RBF, data_dict, features_list)
clf_RF = RandomForestClassifier()
tester.test_classifier(clf_RF, data_dict, features_list)
clf_SVC = SVC(kernel='linear', max_iter=1000)
tester.test_classifier(clf_SVC, data_dict, features_list)
clf_NB = GaussianNB()
tester.test_classifier(clf_NB, data_dict, features_list)
clf_KNN = KNeighborsClassifier()
tester.test_classifier(clf_KNN, data_dict, features_list)
# ### Task 5: Tune your classifier to achieve better than .3 precision and recall
# ### using our testing script. Check the tester.py script in the final project
# ### folder for details on the evaluation method, especially the test_classifier
# ### function. Because of the small size of the dataset, the script uses
tree_features = {}
# show which features correspond to average importances
for idx, elem in enumerate(analyzed_features_list[1:]):
tree_features[elem] = get[idx]
print tree_features
fin_feat_tree = [
'poi', '%frompoi', 'shared_receipt_with_poi', 'exercised_stock_options',
'expenses'
]
#Gaussian NB comparison
clf = GaussianNB()
test_classifier(clf, my_dataset, fin_feat_kbest, folds=1000)
clf = GaussianNB()
test_classifier(clf, my_dataset, fin_feat_tree, folds=1000)
# preparing ground for DT classifier
data = featureFormat(my_dataset, fin_feat_kbest, sort_keys=True)
data2 = featureFormat(my_dataset, fin_feat_tree, sort_keys=True)
labels, features = targetFeatureSplit(data)
labels2, features2 = targetFeatureSplit(data2)
tuned_parameters_tree = [{
'max_depth': [2, 3, 4, 5, 6],
'min_samples_leaf': [1, 2, 3, 4]
}]
# grid_search.fit(features,labels)
# pprint.pprint(grid_search.grid_scores_)
# use K-best to rank the best features
k_best = SelectKBest()
k_best.fit(features, labels)
results_list = zip(k_best.get_support(), features_list[1:], k_best.scores_)
results_list = sorted(results_list, key=lambda x: x[2], reverse=True)
# print the scores for each feature
pprint.pprint(results_list)
# use feature_importances_ from a decision tree classifier to rank the best features
from tester import test_classifier, dump_classifier_and_data
from sklearn import tree
clf_test = tree.DecisionTreeClassifier()
test_classifier(clf_test, my_dataset, features_list)
importance = clf_test.feature_importances_
for i in range(len(importance)):
print features_list[i + 1] + ": " + str(importance[i])
# Using the ranking from K-best anf merging it with the more important features obtained from de Decision tree clasifier
# I decided to select the best of both sets:
features_list = [
'poi', 'exercised_stock_options', 'total_stock_value', 'salary',
'fraction_to_poi', 'restricted_stock', 'shared_receipt_with_poi'
]
### Task 4: Try a varity of classifiers
### Please name your classifier clf for easy export below.
### Note that if you want to do PCA or other multi-stage operations,
### you'll need to use Pipelines. For more info:
labels, features = targetFeatureSplit(data) # In[7]: ### Task 4: Try a varity of classifiers ### Please name your classifier clf for easy export below. ### Note that if you want to do PCA or other multi-stage operations, ### you'll need to use Pipelines. For more info: ### http://scikit-learn.org/stable/modules/pipeline.html # Provided to give you a starting point. Try a variety of classifiers. clf = GaussianNB() test_classifier(clf,my_dataset,features_list) clf1=tree.DecisionTreeClassifier() test_classifier(clf1,my_dataset,features_list) clf2 = AdaBoostClassifier() test_classifier(clf2,my_dataset,features_list) clf3=KNeighborsClassifier(n_neighbors = 4) test_classifier(clf3,my_dataset,features_list) # In[8]: from sklearn.neighbors.nearest_centroid import NearestCentroid clf4 = NearestCentroid() test_classifier(clf4,my_dataset,features_list)
]))
feature_list_default.insert(
0, feature_list_default.pop(feature_list_default.index("poi")))
# test different classifiers with the default feature set (for more details, please see the notebook)
for clf in [
GaussianNB(),
KMeans(),
LogisticRegression(class_weight="balanced"),
SVC(class_weight="balanced"),
ADA(),
DT(class_weight="balanced"),
RF(class_weight="balanced")
]:
tester.test_classifier(clf, final_dataset, feature_list_default)
### 4.1: Evaluate the impact of feature engineering on classification performance (for more details, please refer to the notebook). Test only with SVC (the best classifier).
print "BASELINE PERFORMANCE -default feature set and SVC with linear kernel"
print "--------------------------------------------------------------------"
clf = SVC(kernel="linear", class_weight="balanced")
tester.test_classifier(clf, final_dataset, feature_list_default)
print "EXTENDED FEATURE SET1 PERFORMANCE -default feature set + TF-IDF features and SVC with linear kernel"
print "--------------------------------------------------------------------"
selected_feature_list = feature_list_default + [
'word_feature_2', 'word_feature_3'
]
clf = SVC(kernel="linear", class_weight="balanced")
tester.test_classifier(clf, final_dataset, selected_feature_list)
parameters_NB = dict(SelectKBest__k=range(1, 10))
pipeline = sklearn.pipeline.Pipeline(steps_NB)
grid = GridSearchCV(pipeline, param_grid=parameters_NB, cv=cv, scoring='f1')
grid.fit(features_train, labels_train)
predict = grid.predict(features_test)
report = classification_report(labels_test, predict)
best_params = grid.best_params_
#print report
print "PARAMETERS USED:"
print best_params
print grid.best_score_
clf_GNB = grid.best_estimator_
print "TUNED CLASSIFICATION REPORT:"
test_classifier(clf_GNB, my_dataset, RF_features_list, folds=1000)
#overwrite features_list
features_list = RF_features_list
#tuned Random Forest **without** SKB
steps_RF = [
('minmax', mms),
#('SelectKBest', skb),
('random_forest', clf_RF)
]
parameters_RF = dict( #SelectKBest__k = [6],
random_forest__criterion=['gini'],
random_forest__n_estimators=[9],
random_forest__min_samples_split=[2],
#效果不好，予以去除
#ab_pca = {"PCA__n_components": range(4, 7), "PCA__whiten": [True, False]}
#ab_k.update(ab_pca)
ab_k.update(ab_params)
enron.get_best_parameters_reports(pipe_ab, ab_k, features, labels)
if __name__ == '__main__':
''' GAUSSIAN NAIVE BAYES '''
#设置需要使用的分类器
clf = GaussianNB()
#CrossValidation进行测试
print "Gaussian Naive Bayes : \n", tester.test_classifier(
clf, my_dataset, best_features_list)
"""
Gaussian Naive Bayes :
GaussianNB(priors=None)
Accuracy: 0.84380 Precision: 0.40058 Recall: 0.34550 F1: 0.37101 F2: 0.35527
Total predictions: 15000 True positives: 691 False positives: 1034 False negatives: 1309 True negatives: 11966
None
"""
''' LOGISTIC REGRESSION '''
#使用tune获取每个算法的最佳参数
#tune_logistic_regression()
"""
### Extract features and labels from dataset for local testing data = featureFormat(my_dataset, features_list, sort_keys = True) labels, features = targetFeatureSplit(data) ### Task 4: Try a varity of classifiers ### Please name your classifier clf for easy export below. ### Note that if you want to do PCA or other multi-stage operations, ### you'll need to use Pipelines. For more info: ### http://scikit-learn.org/stable/modules/pipeline.html # Provided to give you a starting point. Try a variety of classifiers. #GaussianNB from sklearn.naive_bayes import GaussianNB clf = GaussianNB() test_classifier(clf,my_dataset,features_list,folds = 1000) #decision tree from sklearn import tree clf = tree.DecisionTreeClassifier(min_samples_leaf=1) test_classifier(clf,my_dataset,features_list,folds = 1000) #Adaboost from sklearn.ensemble import AdaBoostClassifier clf = AdaBoostClassifier() test_classifier(clf,my_dataset,features_list,folds = 1000) #kNearestNeighbours from sklearn.neighbors import KNeighborsClassifier clf=KNeighborsClassifier(n_neighbors = 4) test_classifier(clf,my_dataset,features_list)
for i, clf in enumerate(classifiers):
print 'Step 1: ', i, names[i]
clf.fit(features_train, labels_train)
pred = clf.predict(features_test)
print "Precision: ", precision_score(labels_test, pred)
print "Recall: ", recall_score(labels_test, pred)
print "F1: ", f1_score(labels_test, pred)
print 'done...'
for clf in classifiers:
test_classifier(clf, my_dataset, features_list)
print '-------------------------------------------------------------------------------'
#%% Step 2:
classifier_opt = []
parameters = [
dict(),
dict(n_neighbors=range(1, 20, 1), weights=['uniform', 'distance']),
dict(criterion=['gini', 'entropy'],
min_samples_split=range(10, 30, 1),
min_samples_leaf=range(1, 11, 1)),
dict(criterion=['gini', 'entropy'],
n_estimators=[5, 8, 10, 12, 25],
precision_knn = []
recall_knn = []
# Apply SelectKBest to each classifier for k=1 to k=19
for i in range(1, 20):
k = SelectKBest(f_classif, k=i)
features_new = k.fit_transform(features, labels)
selected_features_index = k.get_support()
selected_features_list = features_list[selected_features_index]
selected_features_list = np.insert(selected_features_list, 0, 'poi')
print "===================="
print "Selected features: ", selected_features_list
# DecisionTree
clf = tree.DecisionTreeClassifier()
pre, rec = test_classifier(clf, my_dataset, selected_features_list, folds = 1000)
precision_tree.append(pre)
recall_tree.append(rec)
# Naive Bayes
clf = naive_bayes.GaussianNB()
pre, rec = test_classifier(clf, my_dataset, selected_features_list, folds = 1000)
precision_nb.append(pre)
recall_nb.append(rec)
# K Nearest Neighbors
clf = neighbors.KNeighborsClassifier(n_neighbors=3)
pre, rec = test_classifier(clf, my_dataset, selected_features_list, folds = 1000)
precision_knn.append(pre)
recall_knn.append(rec)
# Out of the three SVC seems to be most accurate.
# In[89]:
from sklearn.tree import DecisionTreeClassifier
from tester import test_classifier
from sklearn import preprocessing
from sklearn.model_selection import GridSearchCV
from sklearn.pipeline import Pipeline
from sklearn.decomposition import PCA
from sklearn.cross_validation import StratifiedShuffleSplit
test_classifier(DecisionTreeClassifier( random_state = 1), enron_data, features_final, folds = 100)
tree = DecisionTreeClassifier()
parameters = {'tree__criterion': ('gini','entropy'),
'tree__splitter':('best','random'),
'tree__min_samples_split':[2, 10, 20],
'tree__max_depth':[10,15,20,25,30],
'tree__max_leaf_nodes':[5,10,30]}
# use scaling in GridSearchCV
Min_Max_scaler = preprocessing.MinMaxScaler()
#features = Min_Max_scaler.fit_transform(features)
pipeline = Pipeline(steps=[('scaler', Min_Max_scaler), ('pca',PCA(n_components = 2)), ('tree', tree)])
cv = StratifiedShuffleSplit(target, 100, random_state = 42)
('NB', nb_Clf), ('SVM', svm_Clf)]
results = []
names = []
scoring = 'accuracy'
df1 = df
f_list = df.columns
f_list = f_list[1:len(f_list) - 1]
f_list = map(str, f_list)
f_features = ['poi']
for fn in f_list:
f_features.append(fn)
print(f_features)
test_classifier(l_Clf, my_dataset, f_features, folds=45)
test_classifier(lda_Clf, my_dataset, f_features, folds=45)
test_classifier(knn_Clf, my_dataset, f_features, folds=45)
test_classifier(rf_Clf, my_dataset, f_features, folds=45)
print("\n\n")
df['salary_bonus_ratio'] = df.salary.div(df.bonus)
df.loc[~np.isfinite(df['salary_bonus_ratio']), 'salary_bonus_ratio'] = 0
df['salary_expense_ratio'] = df.salary.div(df.expenses)
df.loc[~np.isfinite(df['salary_expense_ratio']), 'salary_expense_ratio'] = 0
features_list.append('salary_bonus_ratio')
features_list.append('salary_expense_ratio')
df = df[features_list]
df = df.apply(np.sqrt, axis=1)
sd = StandardScaler()
fsl = FeatureSel(k_best=5, pca_comp=5)
# clf=Pipeline([("fsl",fsl),("sd",sd),("lvc",LinearSVC(C=0.000001))])
clf = Pipeline([("fsl", fsl), ("sd", sd), ("lvc", LinearSVC())])
gscv=GridSearchCV(clf,{"lvc__C":np.logspace(-6,-1,5),
"fsl__k_best":[1,5,10],
"fsl__pca_comp":[0,5,10]},
scoring="recall",verbose=0)
gscv.fit(np.array(features),np.array(labels))
### Task 5: Tune your classifier to achieve better than .3 precision and recall
### using our testing script.
### Because of the small size of the dataset, the script uses stratified
### shuffle split cross validation. For more info:
### http://scikit-learn.org/stable/modules/generated/sklearn.cross_validation.StratifiedShuffleSplit.html
test_classifier(gscv.best_estimator_, my_dataset, features_list)
### Dump your classifier, dataset, and features_list so
### anyone can run/check your results.
dump_classifier_and_data(gscv.best_estimator_, my_dataset, features_list)
sss = StratifiedShuffleSplit(labels, n_iter =100, test_size=0.3, random_state = 42) grid_search = GridSearchCV(pipeline, param_grid=parameters, cv = sss) ### Tried different parameter for StratifiedShuffleSplit and GridSearcgCV #sss= StratifiedShuffleSplit(n_iter = 20,test_size=0.5, random_state = 5) #grid_search = GridSearchCV(pipeline, param_grid=param_grid, cv = sss, verbose=10, scoring='f1') #grid_search = GridSearchCV(pipeline, param_grid=parameters, cv = sss, error_score = 0, scoring='f1') #print "Grid Search: ", grid_search #print(grid_search.best_estimator_.steps) #print "\n", "Best parameters are: ", grid_search.best_params_, "\n" grid_search.fit(features, labels) clf = grid_search.best_estimator_ ### Use test_classifier.py to test the best model found from tester import test_classifier # Use test_classifier to evaluate the model selected by GridSearchCV print "\n", "Tester Classification report - StratifiedShuffleSplit:" test_classifier(clf, data_dict, features_list) ### Task 6: Dump your classifier, dataset, and features_list so anyone can ### check your results. You do not need to change anything below, but make sure ### that the version of poi_id.py that you submit can be run on its own and ### generates the necessary .pkl files for validating your results. #print features_list dump_classifier_and_data(clf, my_dataset, features_list)
print("Training time : {}".format(end_fitting - start_fitting))
start_predicting = time()
svc_pred = svc_grid.predict(features_test)
end_predicting = time()
print("Predicting time : {}".format(end_predicting - start_predicting))
svc_accuracy = accuracy_score(svc_pred, labels_test)
print('SVC accuracy score : {}'.format(svc_accuracy))
print "f1 score :", f1_score(svc_pred, labels_test)
print "precision score :", precision_score(svc_pred, labels_test)
print "recall score :", recall_score(svc_pred, labels_test)
svc_best_estimator = svc_grid.best_estimator_
print(svc_best_estimator)
test_classifier(nb_grid.best_estimator_, my_dataset, features_list)
#Checking the affect of new feature on the final classifier
test_features_list = [
'poi', 'total_stock_value', 'exercised_stock_options', 'bonus',
'deferred_income', 'long_term_incentive', 'restricted_stock', 'salary',
'total_payments', 'other', 'shared_receipt_with_poi',
'fraction_from_this_person_to_poi'
]
print "\n=================Effect of new feature on final classifier================="
test_classifier(nb_grid.best_estimator_, my_dataset, test_features_list)
###Task 6: Dump your classifier, dataset, and features_list so anyone can
}
dtc_clf = sklearn.tree.DecisionTreeClassifier()
dtcclf = grid_search.GridSearchCV(dtc_clf, parameters, scoring=scoring, cv=cv)
dtcclf.fit(features, labels)
print 'best estimator:', dtcclf.best_estimator_
print 'best score:', dtcclf.best_score_
print 'Processing time:', round(time() - t0, 3), 's'
#Validation of ClassifierClassifier validation
##DecisionTreeClassifier Validation No. 1 (StratifiedShuffleSplit, folds = 1000)
t0 = time()
dtc_best_clf = dtcclf.best_estimator_
test_classifier(dtc_best_clf, enron_data, eng_feature_list)
print 'Processing time:', round(time() - t0, 3), 's'
##DecisionTreeClassifier Validation No. 2 (Randomized, partitioned trials, n=1,000)
t0 = time()
dtc_best_clf = dtcclf.best_estimator_
evaluate.evaluate_clf(dtc_best_clf,
features,
labels,
num_iters=1000,
test_size=0.3)
print("Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
print 'Processing time:', round(time() - t0, 3), 's'
# Importing the AdaBoost from Scikit learn package.
from sklearn import tree
# Creating a classifier with the optimized parameters.
clf_ada = tree.DecisionTreeClassifier(splitter='best',
criterion='gini',
class_weight='balanced',
min_samples_leaf=1,
min_samples_split=2,
max_depth=5,
max_leaf_nodes=4)
# Testing the classifier.
tester.test_classifier(clf=clf_ada,
dataset=my_dataset,
feature_list=features_list)
### Task 5: Tune your classifier to achieve better than .3 precision and recall
### using our testing script. Check the tester.py script in the final project
### folder for details on the evaluation method, especially the test_classifier
### function. Because of the small size of the dataset, the script uses
### stratified shuffle split cross validation. For more info:
### http://scikit-learn.org/stable/modules/generated/sklearn.cross_validation.StratifiedShuffleSplit.html
# Please, find more information in the Jupyter Notebook.
# Example starting point. Try investigating other evaluation techniques!
### Task 6: Dump your classifier, dataset, and features_list so anyone can
### check your results. You do not need to change anything below, but make sure
#
from sklearn.preprocessing import MinMaxScaler
from sklearn.naive_bayes import GaussianNB
from sklearn import linear_model
from sklearn.cross_validation import train_test_split
from sklearn.grid_search import GridSearchCV
from sklearn.feature_selection import SelectKBest, f_classif
from sklearn.svm import SVC
features_train_zero, features_test_zero, labels_train_zero, labels_test_zero = train_test_split(
features, labels, test_size=0.3, random_state=42)
clf = GaussianNB()
clf.fit(features_train_zero, labels_train_zero)
print test_classifier(clf, my_dataset, features_list, folds=1000)
#
#### Task 5: Tune your classifier to achieve better than .3 precision and recall
#### using our testing script. Check the tester.py script in the final project
#### folder for details on the evaluation method, especially the test_classifier
#### function. Because of the small size of the dataset, the script uses
#### stratified shuffle split cross validation. For more info:
#### http://scikit-learn.org/stable/modules/generated/sklearn.cross_validation.StratifiedShuffleSplit.html
#
## Example starting point. Try investigating other evaluation techniques!
#from sklearn.cross_validation import train_test_split
#features_train, features_test, labels_train, labels_test = \
# train_test_split(features, labels, test_size=0.3, random_state=42)
#
#### Task 6: Dump your classifier, dataset, and features_list so anyone can
random_state=42,
stratify=labels)
### Task 4: Try a varity of classifiers
### Please name your classifier clf for easy export below.
### Note that if you want to do PCA or other multi-stage operations,
### you'll need to use Pipelines. For more info:
### http://scikit-learn.org/stable/modules/pipeline.html
# Provided to give you a starting point. Try a variety of classifiers.
# clf = DecisionTreeClassifier(min_samples_split=100)
# clf = SVC(C=10.0, gamma=0.001)
# clf = AdaBoostClassifier(DecisionTreeClassifier(min_samples_split=100), algorithm="SAMME")
clf = KNeighborsClassifier(n_neighbors=5, weights="distance", algorithm="auto")
test_classifier(clf, data_dict, selected_features_list)
### Task 5: Tune your classifier to achieve better than .3 precision and recall
### using our testing script. Check the tester.py script in the final project
### folder for details on the evaluation method, especially the test_classifier
### function. Because of the small size of the dataset, the script uses
### stratified shuffle split cross validation. For more info:
### http://scikit-learn.org/stable/modules/generated/sklearn.cross_validation.StratifiedShuffleSplit.html
# # tuned_parameters = {"criterion": ("gini", "entropy"), "max_depth": (None, 1, 2, 5, 7, 10), "min_samples_split": (10, 100, 250)}
# # tuned_parameters = {"C": (10.0, 100.0, 1000.0), "gamma": (1e-3, 1e-4)}
# # tuned_parameters = {"n_estimators": (50, 100, 150, 200), "learning_rate": (1.0, 1.5, 2.0), "algorithm": ("SAMME", "SAMME.R")}
# tuned_parameters = {"n_neighbors": (1, 5, 10, 15), "weights": ("uniform", "distance")}
#
# gs = GridSearchCV(clf, tuned_parameters, cv=10)
# gs.fit(X_train, y_train)
clf_NB = GaussianNB()
parm = {}
clf_NB = Pipeline([('scaler', scaler), ('gnb', clf_NB)])
gs = GridSearchCV(clf_NB, parm)
gs.fit(features_train, labels_train)
clf_NB = gs.best_estimator_
print "\nGaussianNB score:\n", clf_NB.score(features_train, labels_train)
print "GaussianNB score time:", round(time() - t1, 3), "s"
## Test Point
print "\nGaussianNB:\n", test_classifier(clf_NB, my_dataset, features_list)
## 2. Decision Tree Classifier
t2 = time()
parms = {'criterion': ['gini', 'entropy'], \
'min_samples_split': [2, 5, 10, 20], \
'max_depth': [None, 2, 5, 10], \
'splitter': ['random', 'best'], \
'max_leaf_nodes': [None, 5, 10, 20]}
clf_DT = tree.DecisionTreeClassifier()
gs = GridSearchCV(clf_DT, parms)
parameters = {'max_depth': [1,2,3,4,5,6,8,9,10],
'min_samples_split':[2,3,4,5,6,7,8],
'min_samples_leaf':[1,2,3,4,5,6,7,8,9,10],
'criterion':('gini', 'entropy')}
dt_clf = DecisionTreeClassifier(random_state = 42)
cv = cross_validation.StratifiedShuffleSplit(labels, n_iter=10)
clf = GridSearchCV(dt_clf, parameters,cv=cv, scoring = 'f1')
clf.fit(features,labels)
predictor = clf.predict(features_test)
dt_best_estimator=clf.best_estimator_
precision = precision_score(labels_test,predictor)
recall = recall_score(labels_test,predictor)
f1_score=f1_score(labels_test,predictor)
print "Best score:%f"%clf.best_score_
print dt_best_estimator
print "processing time:", round(time()-t0, 3), "s"
# Classifier validation
##DecisionTreeClassifier Validation 1 (StratifiedShuffleSplit, folds = 1000)
t0 = time()
test_classifier(dt_best_estimator, my_dataset, my_features_list)
print 'Processing time:', round(time() - t0, 3), 's'
### Task 6: Dump your classifier, dataset, and features_list so anyone can
### check your results. You do not need to change anything below, but make sure
### that the version of poi_id.py that you submit can be run on its own and
### generates the necessary .pkl files for validating your results.
dump_classifier_and_data(dt_best_estimator, my_dataset, my_features_list)
### Create StratifiedKFold
skf = StratifiedKFold(labels_train,random_state=42)
### DecisionTreeClassifier
algo_name = "DecisionTreeClassifier"
print algo_name
from sklearn.tree import DecisionTreeClassifier
pipeline = Pipeline([('scaler', MinMaxScaler()), ('kbest', SelectKBest(f_classif)), ('dtc', DecisionTreeClassifier(random_state=42))])
grid_search = GridSearchCV(pipeline, {'kbest__k': range(1,16), 'dtc__min_samples_split': [1,2,3], 'dtc__max_depth': [None, 10, 5]},scoring='f1',cv=skf)
grid_search.fit(features_train, labels_train)
clf = grid_search.best_estimator_
perf_dict[algo_name] = test_classifier(clf, my_dataset, features_list)
parm_dict[algo_name] = grid_search.best_params_
### Print SelectKBest scores, note these are the same for all classifiers
kbest_scores = clf.named_steps['kbest'].scores_
feature_scores = {}
for i in xrange(1,len(features_list)):
feature_scores[features_list[i]] = kbest_scores[i-1]
feature_scores = sorted(feature_scores.items(), key=operator.itemgetter(1),reverse=True)
i = 1
for f in feature_scores:
print "|{}|{}|{}|".format(i,f[0],f[1])
i += 1
### http://scikit-learn.org/stable/modules/pipeline.html
# Provided to give you a starting point. Try a variety of classifiers.
### Extract features and labels from my_dataset for local testing
data = featureFormat(my_dataset, features_list, sort_keys=True)
labels, features = targetFeatureSplit(data)
folds = 1000
cv = StratifiedShuffleSplit(labels, folds, random_state=42)
## Setting up 3 classifiers with feature scaling
## Gaussian NB
print "Gaussian NB classifier output:"
NB_clf = Pipeline(
steps=[('scaling',
preprocessing.MinMaxScaler()), ('classifier', GaussianNB())])
t0 = time()
tester.test_classifier(NB_clf, my_dataset, features_list)
print "Gaussian NB run time:", round(time() - t0, 3), "s"
## KMeans
print "KMeans classifier output:"
KM_clf = Pipeline(
steps=[('scaling',
preprocessing.MinMaxScaler()), ('classifier',
KMeans(n_clusters=2))])
t0 = time()
tester.test_classifier(KM_clf, my_dataset, features_list)
print "KMeans run time:", round(time() - t0, 3), "s"
## Decision tree
print "Decision Tree classifier output:"
DT_clf = Pipeline(steps=[(
#### keep the engineered features added to data_dict
my_dataset = data_dict
### Task 4: Try a varity of classifiers
### Please name your classifier clf for easy export below.
### Note that if you want to do PCA or other multi-stage operations,
### you'll need to use Pipelines. For more info:
### http://scikit-learn.org/stable/modules/pipeline.html
##### Random Forest
from sklearn.ensemble import RandomForestClassifier
rf = RandomForestClassifier(class_weight='auto', random_state=42)
from time import time
from tester import test_classifier
t0 = time()
test_classifier(rf, data_dict, features_list, folds = 100)
print("Random forest fitting time: %rs" % round(time()-t0, 3))
###### Adaboost
from sklearn.ensemble import AdaBoostClassifier
ab = AdaBoostClassifier(random_state=42)
t0 = time()
test_classifier(ab, data_dict, features_list, folds = 100)
print("AdaBoost fitting time: %rs" % round(time()-t0, 3))
### Task 5: Tune your classifier to achieve better than .3 precision and recall
### using our testing script. Check the tester.py script in the final project
### folder for details on the evaluation method, especially the test_classifier
### function. Because of the small size of the dataset, the script uses
### stratified shuffle split cross validation. For more info:
### http://scikit-learn.org/stable/modules/generated/sklearn.cross_validation.StratifiedShuffleSplit.html
'metric': ['manhattan', 'minkowski', 'euclidean'],
'weights': ['distance', 'uniform']
},
cv=cv,
scoring='f1')
knn.fit(features, labels)
print 'K Nearest Neighbors best estimator: ', knn.best_estimator_
print 'K Nearest Neighbors best parameters: ', knn.best_params_
print 'K Nearest Neighbors best score: ', knn.best_score_
# tester.test_classifier(knn.best_estimator_, my_dataset, best_features)
# Pipeline
print "Pipelining..."
pipeline = Pipeline([('normalization', scaler),
('classifier', knn.best_estimator_)])
tester.test_classifier(pipeline, my_dataset, best_features)
# Tune K Means
kmeans = GridSearchCV(
KMeans(),
param_grid={
'n_clusters': [2],
'tol': [0.00000001, 0.0000001, 0.000001, 0.00001, 0.0001, 0.001, 0.01],
'max_iter': [300, 200, 400, 500, 600, 700],
'init': ['k-means++', 'random'],
'copy_x': [True, False]
},
cv=cv,
scoring='f1')
kmeans.fit(features, labels)
print 'K Means best estimator: ', kmeans.best_estimator_
