def predict_point(y_train, y_test, X_train, X_test, model):
"""
Method to fit a regression on one point, given as (t, x, y)
:return: y vector of len (t) as a regression prediction
"""
if np.count_nonzero(~np.isnan(y_train)) == 0: # if point is empty
pred = np.empty_like(y_test)
pred[:] = np.nan
else:
if model == 'lasso':
model = Lasso(alpha=0.1, max_iter=1000)
y_clean, X_train_clean = clean_data(X=X_train, y=y_train)
model.fit(X=X_train_clean, y=y_clean)
mask = ~np.isnan(X_test).any(axis=1)
X_test_clean = X_test[mask]
pred = model.predict(X_test_clean)
pred_out = np.empty_like(y_test)
pred_out[mask] = pred
pred_out[~mask] = np.nan
pred_out[pred_out < 0] = 0
return pred_out
elif model == 'lstm':
model = Sequential()
model.add(LSTM(50, input_shape=(X_train.shape[1], X_train.shape[2]), return_sequences=False))
model.add(Dense(1))
model.compile(loss='mae', optimizer='adam')
x_train = data_train[feature].as_matrix()
y_train = data_train[v].as_matrix()
from keras.models import Sequential
from keras.layers.core import Dense, Activation
import time
start = time.clock()
# 输入层为３个节点，隐藏层６个节点
model = Sequential() # 建立模型
model.add(Dense(output_dim=6, input_dim=8)) # 添加输入层、隐藏层节点
# model.add(Dense(input_dim=8,12))
model.add(Activation('relu')) # 使用relu作为激活函数，可以大幅度提高准确率
model.add(Dense(units=1, input_dim=8)) # 添加输出层节点
model.compile(loss='mean_squared_error', optimizer='adam') # 编译模型
model.fit(x_train, y_train, nb_epoch=3000, batch_size=16) #训练模型，学习一千次
end = time.clock()
usetime = end - start
print('训练该模型耗时' + str(usetime) + 's!')
aa = v + '.model'
model.save_weights(aa) # 将该模型存储
x = ((fatigue[feature] - data_mean[feature]) /
data_std[feature]).as_matrix()
pre = v + 'pre'
fatigue[pre] = model.predict(x) * data_std[v] + data_mean[v]
#保存的表名命名格式为“2_2_3_1k此表功能名称”，是此小节生成的第1张表格，功能为revenue：增值税预测结果
fatigue.to_excel('2_2_3_1zengzhi.xlsx')
import matplotlib.pyplot as plt