def main():
# load training data
filename_train = './data/train.csv'
train_dataset = transform(filename_train)
X = train_dataset['data']
y = train_dataset['target']
# fill in missing data (optional)
X_full = fill_missing(X, 'mode', False)
X_full_train, X_full_test, y_train, y_test = train_test_split(X_full, y, test_size=0.25, random_state=0)
### use the logistic regression
print('Train the logistic regression classifier')
""" your code here """
lr_model = LogisticRegression()
start_time = time.time()
lr_model.fit(X_full_train,y_train)
elapsed_time = time.time() - start_time
y_predict = lr_model.predict(X_full_test)
print('The accuracy of the sklearn lr classifier: '+str(sum(y_test == y_predict)/y_test.shape[0])+' elapsed time: '+str(elapsed_time))
clf = logisticRegression()
start_time = time.time()
clf.fit(X_full_train,y_train)
elapsed_time = time.time() - start_time
y_predict = clf.predict(X_full_test)
print('The accuracy of my lr classifier: '+str(sum(y_test == y_predict)/y_test.shape[0])+' elapsed time: '+str(elapsed_time))
### use the naive bayes
print('Train the naive bayes classifier')
""" your code here """
nb_model = MultinomialNB()
start_time = time.time()
nb_model.fit(X_full_train, y_train)
elapsed_time = time.time() - start_time
y_predict = nb_model.predict(X_full_test)
print('The accuracy of the sklearn nb classifier: '+str(sum(y_test == y_predict)/y_test.shape[0])+' elapsed time: '+str(elapsed_time))
clf = NaiveBayes()
start_time = time.time()
clf = clf.fit(X_full_train, y_train)
elapsed_time = time.time() - start_time
y_predict = clf.predict(X_full_test)
print('The accuracy of my nb classifier: '+str(sum(y_test == y_predict)/y_test.shape[0])+' elapsed time: '+str(elapsed_time))
## use the svm
print('Train the SVM classifier')
""" your code here """
svm_model = svm.SVC(kernel='linear', C=1).fit(X_full_train, y_train)
print(('The accuracy of the sklearn SVM classifier: %f')%(svm_model.score(X_full_test, y_test)))
## use the random forest
print('Train the random forest classifier')
rf_model = RandomForestClassifier(n_estimators=500)
rf_model.fit(X_full_train, y_train)
print(('The accuracy of the sklearn random forest classifier: %f')%(rf_model.score(X_full_test, y_test)))
## get predictions
df = pd.read_csv('./data/test.csv')
UserID=df.loc[:,'UserID'].as_matrix()
df = df.drop('UserID', 1)
X_predict=df.as_matrix()
for n in range(df.shape[1]):
if df.iloc[:,n].dtypes!=np.int64 and df.iloc[:,n].dtypes!=np.float64:
g= pd.get_dummies(X_predict[:,n])
i=0
for e in list(g):
X_predict[:,n][X_predict[:,n]==e]=i
i=i+1
X_full_predict = fill_missing(X_predict, 'mode', False)
y_predict = lr_model.predict(X_full_predict)
fo = open("./predictions/lr_predictions.csv", "w")
fo.write("UserID,Happy\n");
for i in range(y_predict.shape[0]):
fo.write(("%d,%d\n")%(UserID[i],y_predict[i]));
fo.close()
y_predict = nb_model.predict(X_full_predict)
fo = open("./predictions/nb_predictions.csv", "w")
fo.write("UserID,Happy\n");
for i in range(y_predict.shape[0]):
fo.write(("%d,%d\n")%(UserID[i],y_predict[i]));
fo.close()
y_predict = svm_model.predict(X_full_predict)
fo = open("./predictions/svm_predictions.csv", "w")
fo.write("UserID,Happy\n");
for i in range(y_predict.shape[0]):
fo.write(("%d,%d\n")%(UserID[i],y_predict[i]));
fo.close()
y_predict = rf_model.predict(X_full_predict)
fo = open("./predictions/rf_predictions.csv", "w")
fo.write("UserID,Happy\n");
for i in range(y_predict.shape[0]):
fo.write(("%d,%d\n")%(UserID[i],y_predict[i]));
fo.close()
from matplotlib import pyplot as plt from preprocess import transform from preprocess import fill_missing ## import data my_csv = './data/train.csv' ## path to your dataset filename_train = './data/train.csv' train_dataset = transform(filename_train) X = train_dataset['data'] y = train_dataset['target'] # fill in missing data (optional) dat, discard_row = fill_missing(X, 'most_frequent', False) y = np.delete(y, discard_row) # if no row or column titles in your csv, pass 'header=None' into read_csv # and delete 'index_col=0' -- but your biplot will be clearer with row/col names ## perform PCA n = dat.shape[1] pca = PCA(n_components=n) # defaults number of PCs to number of columns in imported data (ie number of # features), but can be set to any integer less than or equal to that value pca.fit(dat) ## project data into PC space
def main():
# load training data
filename_train = "./data/train.csv"
filename_test = "./data/test.csv"
df = pd.read_csv(filename_test, header=0)
X_pre_userId = df['UserID']
X_pre_userId = X_pre_userId.as_matrix()
train_dataset = transform(filename_train)
test_dateset = transform(filename_test)
X = train_dataset['data']
y = train_dataset['target']
X_pre = test_dateset['data']
num_train = X.shape[0]
X = np.append(X, X_pre, 0)
X_fill = fill_missing(X, 'most_frequent', False)
# X_fill = fill_missing(X, 'most_frequent', True)
X_pre_fill = X_fill[num_train::]
X_fill = X_fill[0:num_train]
X_train, X_test, y_train, y_test = train_test_split(X_fill,
y,
test_size=0.2,
random_state=4)
print(y_train.shape, y_test.shape)
### use the logistic regression
print('Train the logistic regression classifier')
""" your code here """
lr_model = LogisticRegression(random_state=4)
lr_model.fit(X_train, y_train)
print(lr_model.score(X_test, y_test))
lr_pre = lr_model.predict(X_pre_fill)
file = open('./predictions/lr_predictions.csv', 'w')
file.write('UserID,Happy\n')
for temp in range(0, lr_pre.shape[0]):
file.write('%d' % X_pre_userId[temp])
file.write(',')
file.write(str(lr_pre[temp]))
file.write('\n')
### use the naive bayes
print('Train the naive bayes classifier')
""" your code here """
nb_model = GaussianNB()
nb_model.fit(X_train, y_train)
print(nb_model.score(X_test, y_test))
nb_pre = nb_model.predict(X_pre_fill)
file = open('./predictions/nb_predictions.csv', 'w')
file.write('UserID,Happy\n')
for temp in range(0, nb_pre.shape[0]):
file.write('%d' % X_pre_userId[temp])
file.write(',')
file.write(str(nb_pre[temp]))
file.write('\n')
## use the svm
print('Train the SVM classifier')
""" your code here """
svm_model = svm.SVC(kernel='linear', random_state=0)
svm_model.fit(X_train, y_train)
print(svm_model.score(X_test, y_test))
svm_pre = svm_model.predict(X_pre_fill)
file = open('./predictions/svm_predictions.csv', 'w')
file.write('UserID,Happy\n')
for temp in range(0, svm_pre.shape[0]):
file.write('%d' % X_pre_userId[temp])
file.write(',')
file.write(str(svm_pre[temp]))
file.write('\n')
## use the random forest
print('Train the random forest classifier')
""" your code here """
rf_model = RandomForestClassifier(n_estimators=2600, random_state=4)
rf_model = rf_model.fit(X_train, y_train)
print(rf_model.score(X_test, y_test))
rf_pre = rf_model.predict(X_pre_fill)
file = open('./predictions/rf_predictions.csv', 'w')
file.write('UserID,Happy\n')
for temp in range(0, rf_pre.shape[0]):
file.write('%d' % X_pre_userId[temp])
file.write(',')
file.write(str(rf_pre[temp]))
file.write('\n')
## get predictions
""" your code here """
def main():
# load training data
filename_train = './data/train.csv'
train_dataset = transform(filename_train)
X = train_dataset['data']
y = train_dataset['target']
# fill in missing data (optional)
X_full, discard_row = fill_missing(X, 'most_frequent', True)
y = np.delete(y,discard_row)
n_samples, n_features = X_full.shape
### -------------------- use the logistic regression --------------------
print('\n\nTrain the logistic regression classifier')
train_X, train_y, valid_X, valid_y = cross_validation(0.08,X_full,y) #0.08
# Sklearn package
lr_model_time1 = time.time()
lr_model = LogisticRegression()
lr_model = lr_model.fit(train_X,train_y)
lr_model_time2 = time.time()
print("Sklearn LR validation score: {0}".format(lr_model.score(valid_X,valid_y)))
print("Sklearn LR training time: %.3f s" % (lr_model_time2 - lr_model_time1))
#print("Sklearn LR learnt coef:\n{0},\n{1}".format(lr_model.coef_[:,:5],lr_model.intercept_))
# Self-implemented
train_X, train_y, valid_X, valid_y = cross_validation(0.15,X_full,y) #0.15
self_lr_time1 = time.time()
self_lr = LogitR()
self_lr = self_lr.fit(train_X,train_y)
self_lr_time2 = time.time()
print("Self LR validation score: {0}".format(self_lr.score(valid_X,valid_y)))
print("Self LR training time: %.3f s" % (self_lr_time2 - self_lr_time1))
#print("Self LR learnt coef:\n{0},\n{1}".format(self_lr.coef[:5],self_lr.intercept))
### -------------------- use the logistic regression --------------------
### -------------------- use the naive bayes --------------------
# Sklearn package
print('\n\nTrain the naive bayes classifier')
train_X, train_y, valid_X, valid_y = cross_validation(0.1,X_full,y) # Sklearn NB validation score: 0.6762589928057554
nb_model_time1 = time.time()
nb_model = BernoulliNB()
nb_model.fit(train_X,train_y)
nb_model_time2 = time.time()
print("Sklearn NB validation score: {0}".format(nb_model.score(valid_X,valid_y)))
print("SKlearn NB training time: %.3f s" % (nb_model_time2 - nb_model_time1))
#sk_y_predict = nb_model.predict(X_full[1800:,1:n_features-1])
# Self-implemented
train_X, train_y, valid_X, valid_y = cross_validation(0.118,X_full,y) # Self NB validation score: 0.576 # i 0.118
self_nb_time1 = time.time()
self_nb = NaiveBayes()
self_nb = self_nb.fit(train_X,train_y)
self_nb_time2 = time.time()
print("Self NB validation score: {0}".format(self_nb.score(train_X,train_y)))
print("Self NB training time: %.3f s" % (self_nb_time2 - self_nb_time1))
#self_y_predict = clf.predict(X_full[1800:,1:n_features-1])
### -------------------- use the naive bayes --------------------
### -------------------- use svm --------------------
print('\n\nTrain the SVM classifier')
# linear, poly, rbf, or precomputed (or self-defined)?
train_X, train_y, valid_X, valid_y = cross_validation(0.17,X_full,y) #0.17
svm_model_time1 = time.time()
svm_model = svm.SVC(kernel="linear")
# rbf score: 0.682; validation percentage: 0.113
# sigmoid score: 0.577; validation percentage: 0.23
# poly score: 0.685; validation percentage: 0.16
# linear score: 0.701 validation percentage: 0.17
svm_model.fit(train_X,train_y)
print("train_X:", train_X.shape)
print("train_y:", train_y.shape)
svm_model_time2 = time.time()
print("Sklearn SVM validation score: {0}".format(svm_model.score(valid_X,valid_y)))
print("Sklearn SVM training time: %.3f s" % (svm_model_time2 - svm_model_time1))
### -------------------- use svm --------------------
### -------------------- use random forest --------------------
print('\n\nTrain the random forest classifier')
train_X, train_y, valid_X, valid_y = cross_validation(0.151,X_full,y) # Sklearn RF validation score: 0.702 # i: 0.151
rf_model_time1 = time.time()
rf_model = RandomForestClassifier(n_estimators=29) # 29
rf_model.fit(train_X,train_y)
rf_model_time2 = time.time()
print("Sklearn RF validation score: {0}".format(rf_model.score(valid_X,valid_y)))
print("Sklearn RF training time: %.3f s" % (rf_model_time2 - rf_model_time1))
### -------------------- use random forest --------------------
## get predictions
""" your code here """
income_id = 3
educate_id = 10
h = 0.2
filename_train = '../data/train.csv'
train_dataset = transform(filename_train)
X = train_dataset['data']
y = train_dataset['target']
'''
row_idx = X[:,0]
mat = X[:,income_id+1:educate_id+2:educate_id-income_id]
col_name = ['Income', 'EducationLevel']
X_show = pd.DataFrame(data=mat, index=row_idx, columns=col_name)
'''
X_full, discard_row = fill_missing(X, 'most_frequent', True)
#X_full = X_full[:,1:X_full.shape[1]-1]
X_full = X_full[:, income_id:educate_id + 1:educate_id - income_id]
y_full = np.delete(y, discard_row)
#x_min, x_max = min(X_full[:,income_id])-1, max(X_full[:,income_id])+1
#y_min, y_max = min(X_full[:,educate_id])-1, max(X_full[:,educate_id])+1
x_min, x_max = min(X_full[:, 0]) - 1, max(X_full[:, 0]) + 1
y_min, y_max = min(X_full[:, 1]) - 1, max(X_full[:, 1]) + 1
xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h))
svm_model = svm.SVC(kernel="rbf")
#svm_model = svm.LinearSVC(C=1)
svm_model.fit(X_full, y_full)
print("SVM end")
data[:, n][data[:, n] == 'Domestic Partners (w/kids)'] = 5
n = df.columns.get_loc("EducationLevel")
data[:, n][data[:, n] == 'Current K-12'] = 0
data[:, n][data[:, n] == 'High School Diploma'] = 1
data[:, n][data[:, n] == 'Current Undergraduate'] = 2
data[:, n][data[:, n] == "Associate's Degree"] = 3
data[:, n][data[:, n] == "Bachelor's Degree"] = 4
data[:, n][data[:, n] == "Master's Degree"] = 5
data[:, n][data[:, n] == 'Doctoral Degree'] = 6
for n in range(df.shape[1]):
if df.iloc[:, n].dtypes != np.int64 and df.iloc[:, n].dtypes != np.float64:
g = pd.get_dummies(data[:, n])
i = 0
for e in list(g):
data[:, n][data[:, n] == e] = i
i = i + 1
X_full = fill_missing(data, 'mode', False)
mins = np.min(X_full, axis=0)
maxs = np.max(X_full, axis=0)
X_full = (X_full - np.mean(X_full, axis=0)) / (maxs - mins)
pca = PCA()
X_reduced = pca.fit_transform(X_full)
fig = plt.figure('PCA and biplot')
biplot(X_reduced, pca.components_, 1, 2, list(df))
plt.show()
fig.savefig('PCA and biplot.jpg')
filename = '../data/train.csv'
dataset = preprocess.transform(filename)
#dataset = preprocess.fill_missing(dataset,strategy = 'most_frequent',isClassified = False)
X = dataset['data']
y = dataset['target']
# drop NaN
total = (pd.concat([X, y], axis=1))
total = total.dropna()
#X = total.drop('Happy',1)
total = total[['Income', 'EducationLevel', 'Happy']].dropna()
total = preprocess.fill_missing(total,
strategy='most_frequent',
isClassified=False)
y = total['Happy']
X = total[['Income', 'EducationLevel']]
X = np.array(X)
y = np.array(y)
# train svm model
'''
svm_model = svm.SVC(kernel='rbf')
svm_model = svm_model.fit(X,y)
y_predict_svm = svm_model.predict(X)
'''
C = 1.0 # SVM regularization parameter
data[:, n][data[:, n] == '$50,000 - $74,999'] = 2
data[:, n][data[:, n] == '$75,000 - $100,000'] = 3
data[:, n][data[:, n] == '$100,001 - $150,000'] = 4
data[:, n][data[:, n] == 'over $150,000'] = 5
n = df.columns.get_loc("EducationLevel")
data[:, n][data[:, n] == 'Current K-12'] = 0
data[:, n][data[:, n] == 'High School Diploma'] = 1
data[:, n][data[:, n] == 'Current Undergraduate'] = 2
data[:, n][data[:, n] == "Associate's Degree"] = 3
data[:, n][data[:, n] == "Bachelor's Degree"] = 4
data[:, n][data[:, n] == "Master's Degree"] = 5
data[:, n][data[:, n] == 'Doctoral Degree'] = 6
X = data
X = fill_missing(X, 'mode', False)
h = .2 # step size in the mesh
# we create an instance of SVM and fit out data. We do not scale our
# data since we want to plot the support vectors
C = 1.0 # SVM regularization parameter
svc = svm.LinearSVC(C=C).fit(X, y)
#svc = svm.LinearSVC(C=C).fit(X, y)
# create a mesh to plot in
x_min, x_max = X[:, 0].min() - 1, X[:, 0].max() + 1
y_min, y_max = X[:, 1].min() - 1, X[:, 1].max() + 1
xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h))
Z = svc.predict(np.c_[xx.ravel(), yy.ravel()])
# Put the result into a color plot