In [1]:

    

%matplotlib inline
import numpy as np
import pandas as pd
import tensorflow as tf
from matplotlib import pyplot as plt

from tensorflow.contrib import learn
from sklearn.metrics import mean_squared_error

from lstm import x_sin, sin_cos, generate_data, lstm_model


    

In [2]:

    

LOG_DIR = './ops_logs'
TIMESTEPS = 5
RNN_LAYERS = [{'steps': TIMESTEPS}]
DENSE_LAYERS = None
TRAINING_STEPS = 10000
BATCH_SIZE = 100
PRINT_STEPS = TRAINING_STEPS / 100


    

In [3]:

    

regressor = learn.TensorFlowEstimator(model_fn=lstm_model(TIMESTEPS, RNN_LAYERS, DENSE_LAYERS), n_classes=0,
                                      verbose=2,  steps=TRAINING_STEPS, optimizer='Adagrad',
                                      learning_rate=0.03, batch_size=BATCH_SIZE)


    

In [4]:

    

X, y = generate_data(np.sin, np.linspace(0, 100, 10000), TIMESTEPS, seperate=False)
# create a lstm instance and validation monitor
validation_monitor = learn.monitors.ValidationMonitor(X['val'], y['val'],
                                                      every_n_steps=PRINT_STEPS,
                                                      early_stopping_rounds=1000)
regressor.fit(X['train'], y['train'], monitors=[validation_monitor], logdir=LOG_DIR)


    

    
Out[4]:





TensorFlowEstimator(continue_training=False, optimizer=Adagrad, verbose=2, learning_rate=0.03, batch_size=100, steps=10000, n_classes=0, params=None, clip_gradients=5.0, class_weight=None)

In [5]:

    

predicted = regressor.predict(X['test'])
rmse = np.sqrt(((predicted - y['test']) ** 2).mean(axis=0))
score = mean_squared_error(predicted, y['test'])
print ("MSE: %f" % score)


    

    




MSE: 0.000105

In [6]:

    

plot_predicted, = plt.plot(predicted, label='predicted')
plot_test, = plt.plot(y['test'], label='test')
plt.legend(handles=[plot_predicted, plot_test])


    

    
Out[6]:





<matplotlib.legend.Legend at 0x11b354b90>






    

In [23]:

    

LOG_DIR = './ops_logs'
TIMESTEPS = 10
RNN_LAYERS = [{'steps': TIMESTEPS}]
DENSE_LAYERS = None
TRAINING_STEPS = 10000
BATCH_SIZE = 100
PRINT_STEPS = TRAINING_STEPS / 100

regressor = learn.TensorFlowEstimator(model_fn=lstm_model(TIMESTEPS, RNN_LAYERS, DENSE_LAYERS), n_classes=0,
                                      verbose=1,  steps=TRAINING_STEPS, optimizer='Adagrad',
                                      learning_rate=0.03, batch_size=BATCH_SIZE)


    

In [24]:

    

X, y = generate_data(sin_cos, np.linspace(0, 100, 10000), TIMESTEPS, seperate=False)
# create a lstm instance and validation monitor
validation_monitor = learn.monitors.ValidationMonitor(X['val'], y['val'],
                                                      every_n_steps=PRINT_STEPS,
                                                      early_stopping_rounds=1000)
regressor.fit(X['train'], y['train'], monitors=[validation_monitor], logdir=LOG_DIR+'2')


    

    




WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .
WARNING:tensorflow:Input iterator is exhausted: .






    
Out[24]:





TensorFlowEstimator(continue_training=False, optimizer=Adagrad, verbose=1, learning_rate=0.03, batch_size=100, steps=10000, n_classes=0, params=None, clip_gradients=5.0, class_weight=None)

In [25]:

    

predicted = regressor.predict(X['test'])
rmse = np.sqrt(((predicted - y['test']) ** 2).mean(axis=0))
score = mean_squared_error(predicted, y['test'])
print ("MSE: %f" % score)


    

    




MSE: 0.000227

In [26]:

    

plt.plot(predicted)


    

    
Out[26]:





[<matplotlib.lines.Line2D at 0x113b83b10>,
 <matplotlib.lines.Line2D at 0x113b83d90>]






    

In [27]:

    

plt.plot(y['test'])


    

    
Out[27]:





[<matplotlib.lines.Line2D at 0x113281250>,
 <matplotlib.lines.Line2D at 0x1132814d0>]






    

In [37]:

    

plot_predicted = plt.plot(predicted, label='predicted')
plot_test = plt.plot(y['test'], label='test')
plt.legend(handles=plot_predicted+plot_test)


    

    
Out[37]:





<matplotlib.legend.Legend at 0x114647490>






    

In [19]:

    

LOG_DIR = './ops_logs'
TIMESTEPS = 40
RNN_LAYERS = [{'steps': TIMESTEPS}]
DENSE_LAYERS = [2]
TRAINING_STEPS = 130000
BATCH_SIZE = 100
PRINT_STEPS = TRAINING_STEPS / 100

regressor = learn.TensorFlowEstimator(model_fn=lstm_model(TIMESTEPS, RNN_LAYERS, DENSE_LAYERS), n_classes=0,
                                      verbose=1,  steps=TRAINING_STEPS, optimizer='Adagrad',
                                      learning_rate=0.03, batch_size=BATCH_SIZE)


    

In [ ]:

    

X, y = generate_data(x_sin, np.linspace(0, 100, 10000), TIMESTEPS, seperate=False)


    

In [ ]:

    

# create a lstm instance and validation monitor
validation_monitor = learn.monitors.ValidationMonitor(X['val'], y['val'], n_classes=0,
                                                      every_n_steps=PRINT_STEPS,
                                                      early_stopping_rounds=1000)
regressor.fit(X['train'], y['train'], monitors=[validation_monitor], logdir=LOG_DIR)


    

    




WARNING:tensorflow:<tensorflow.python.ops.rnn_cell.BasicLSTMCell object at 0x124708950>: Using a concatenated state is slower and will soon be deprecated.  Use state_is_tuple=True.

In [20]:

    

predicted = regressor.predict(X['test'])
score = mean_squared_error(predicted, y['test'])
print "MSE: {}".format(score)


    

    




MSE: 61.024454351

In [8]:

    

plt.plot(predicted)


    

    
Out[8]:





<matplotlib.axes._subplots.AxesSubplot at 0x11f77d810>






    

In [14]:

    

plt.plot(y['test'])


    

    
Out[14]:





<matplotlib.axes._subplots.AxesSubplot at 0x118452990>






    

In [19]:

    

plt.plot(regressor.predict(X['train']))
plt.plot(y['train'])


    

    
Out[19]:





[<matplotlib.lines.Line2D at 0x1189b8fd0>]






    

In [ ]: