def TrainStacking(df_train1, df_train2, models):
y_stack = df_train2[target].values.astype(float)
X_stack = np.array([])
for dt, mt, m, C in models:
if dt == 'nume':
X_train1 = df_train1[C].values.astype(float)
X_train2 = df_train2[C].values.astype(float)
else:
X_train1 = df_train1[C].values
X_train2 = df_train2[C].values
X_train1, X_train2 = dp.CateEncoderTwoMat(X_train1, X_train2)
y_train = df_train1[target].values.astype(float)
if mt == 'classification':
y_train = np.where(y_train > 0, 1, 0)
m.fit(X_train1, y_train)
p_stack = m.predict(X_train2)
p_stack = np.reshape(p_stack, (len(p_stack), 1))
X_stack = np.hstack((X_stack, p_stack)) if X_stack.size else p_stack
model_stack = LinearRegression()
model_stack.fit(X_stack, y_stack)
# stack_coef = model.coef_
return models, model_stack
def TrainAndValidationClassification(df_train,
df_val,
model,
columns,
data_type='nume'):
sys.stdout.write('Extract data ... ')
sys.stdout.flush()
if data_type == 'nume':
X_train = df_train[columns].values.astype(float)
X_val = df_val[columns].values.astype(float)
else:
X_train = df_train[columns].values
X_val = df_val[columns].values
X_train, X_val = dp.CateEncoderTwoMat(X_train, X_val)
target = 'logerror_sign'
y_train = df_train[target].values.astype(float)
y_val = df_val[target].values.astype(float)
# y_train = abs(y_train)
# y_val = abs(y_val)
sys.stdout.write('done.\n')
sys.stdout.write('Train ... ')
sys.stdout.flush()
model.fit(X_train, y_train)
sys.stdout.write('done.\n')
sys.stdout.write('Predict ...')
sys.stdout.flush()
p_train = model.predict(X_train)
p_val = model.predict(X_val)
sys.stdout.write('done.\n')
acc_train = (p_train == y_train).sum() / float(len(y_train))
acc_val = (p_val == y_val).sum() / float(len(y_val))
print 'Accuracy in train :', acc_train
print 'Accuracy in validation :', acc_val
return y_train, y_val, p_train, p_val
def TrainAndValidation(df_train, df_val, model, columns, data_type='nume'):
sys.stdout.write('Extract data ... ')
sys.stdout.flush()
if data_type == 'nume':
X_train = df_train[columns].values.astype(float)
X_val = df_val[columns].values.astype(float)
else:
X_train = df_train[columns].values
X_val = df_val[columns].values
X_train, X_val = dp.CateEncoderTwoMat(X_train, X_val)
y_train = df_train[target].values.astype(float)
y_val = df_val[target].values.astype(float)
# y_train = abs(y_train)
# y_val = abs(y_val)
sys.stdout.write('done.\n')
sys.stdout.write('Train ... ')
sys.stdout.flush()
model.fit(X_train, y_train)
sys.stdout.write('done.\n')
sys.stdout.write('Predict ...')
sys.stdout.flush()
p_train = model.predict(X_train)
p_val = model.predict(X_val)
sys.stdout.write('done.\n')
mae_train = mean_absolute_error(y_train, p_train)
mae_val = mean_absolute_error(y_val, p_val)
print 'MAE in train :', mae_train
print 'MAE in validation :', mae_val
return y_train, y_val, p_train, p_val