#coding=utf-8
import numpy as np
from pandas import DataFrame, read_csv
import matplotlib.pyplot as plt
#from keras.utils.visualize_util import plot
import DNN
def select_input_datasets(data, y, x,step):
inputs = data[x].values.tolist()
new_input_sets = []
for i in range(len(inputs)):
x = [] # 2D
for j in range(step-1,-1,-1):
if (i-j) < 0:
x.append([0]*len(inputs[0]))
else:
x.append(inputs[i-j])
new_input_sets.append(x)
#print np.array(new_input_sets).shape
return {'y':np.array(data[y].values),'x':np.array(new_input_sets)}
def read_data(train_file_name):
print ("The train data is loading")
return read_csv(train_file_name)
def training(train_data_name=None, time_step=None, x_variables=None, target=None,
dim_hidden_layer=None, lr=None, loss=None, batch_size=None, nb_epoch=None, validation_split=None,
best_model_weight_name=None, last_model_weight_name=None):
original_train_data = read_data(train_data_name)
train_input = select_input_datasets(original_train_data,
y=target,
x=x_variables,
step=time_step)
DNN_model = DNN.DNN(train_input)
# build a DNN model
DNN_model.build_model(dim_hidden_layer=dim_hidden_layer, lr=lr, loss=loss)
DNN_model.shows_model()
# fit the model according to training data
DNN_model.fit_model(batch_size=batch_size, nb_epoch=nb_epoch, validation_split=validation_split, best_model_weight_name=best_model_weight_name)
DNN_model.save_model_to_file(last_model_weight_name)
DNN_model.predict(train_input, show=True)
DNN_model.evaluate(train_input)
return DNN_model
def testing(test_data_name=None, time_step=None, x_variables=None, target=None,
restore_model_weight_name=None):
original_test_data = read_data(test_data_name)
testing_input = select_input_datasets(original_test_data,
y=target,
x=x_variables,
step=time_step)
testing_input['x'] = testing_input['x'][0:]
testing_input['y'] = testing_input['y'][0:]
DNN_model = DNN.DNN(testing_input)
DNN_model.restore_model_from_file(restore_model_weight_name)
DNN_model.predict(testing_input, show=True)
DNN_model.evaluate(testing_input)
return DNN_model
if __name__ == "__main__":
#**************************************************
# training or testing
only_test = False
a='0'
# 'Best_model.weights' or 'last_model'
restore_model_weight_name = 'C:\\Users\\Iron\\Desktop\\0\\best_model.weights'
# data
train_data_name = 'C:\\Users\\Iron\\Desktop\\0\\train_data.csv'
test_data_name = 'C:\\Users\\Iron\\Desktop\\0\\test_data.csv'
target = 'w'
x_variables = [
"a1","a2","a3","a4",
"a6","a6","a7","a8",
"a9","a10"
]
# model param
time_step = 2
dim_hidden_layer = 64
lr = 0.001
loss = 'mape'
# fit param
batch_size = 2
nb_epoch = 15000
validation_split = 0.05
# save
best_model_weight_name = 'C:\\Users\\Iron\\Desktop\\0\\best_model.weights'
last_model_weight_name = 'C:\\Users\\Iron\\Desktop\\0\\last_model.weights'
#**************************************************
if not only_test:
model = training(train_data_name=train_data_name,
time_step=time_step,
x_variables=x_variables, target=target,
dim_hidden_layer=dim_hidden_layer, lr=lr, loss=loss,
batch_size=batch_size, nb_epoch=nb_epoch, validation_split=validation_split,
best_model_weight_name=best_model_weight_name,
last_model_weight_name=last_model_weight_name)
if only_test:
model = testing(test_data_name=test_data_name,
time_step=time_step,
x_variables=x_variables, target=target,
restore_model_weight_name=restore_model_weight_name,)
uj5u.com熱心網友回復：
某個函式需要一個 units=xxx的變數，但是你沒有給入。應該報錯資訊中會有行數。
轉載請註明出處，本文鏈接：https://www.uj5u.com/qita/18214.html