In [1]:

    

import os, sys
sys.path.append(os.path.abspath('../../main/python'))
import thalesians.tsa.neural.tfwrapper as tfwrapper


    

In [2]:

    

%matplotlib inline


    

In [3]:

    

import numpy as np
import matplotlib.pyplot as plt


    

In [4]:

    

x = np.random.normal(size=(100, 5))


    

In [5]:

    

y = np.empty((np.shape(x)[0], 3))


    

In [6]:

    

y[:,0] = np.exp(x[:,0])


    

In [7]:

    

y[:,1] = np.cos(x[:,1]) + np.sin(x[:,2])


    

In [8]:

    

y[:,2] = x[:,0] + x[:,1] + x[:,2] + x[:,3] + x[:,4]


    

In [9]:

    

n, p = np.shape(x)
q = np.shape(y)[1]


    

In [10]:

    

n_train = n // 2
n_test = n - n_train


    

In [11]:

    

x_train = x[:n_train,:]
y_train = y[:n_train,:]
x_test = x[n_train:,:]
y_test = y[n_train:,:]


    

In [12]:

    

model = tfwrapper.Sequential()


    

In [13]:

    

model.add(tfwrapper.Dense(100, input_dim=p, activation='tanh'))
model.add(tfwrapper.Dense(100, activation='tanh'))
model.add(tfwrapper.Dense(q, activation='linear'))


    

In [14]:

    

model.compile(loss='mean_squared_error', optimizer='sgd', lr=0.001)


    

In [15]:

    

y_train_pred = model.fit(x_train, y_train, epochs=10000)


    

    




loss: 0.195759

In [16]:

    

plt.plot(y_train[:,0], y_train_pred[:,0], 'x');


    

In [17]:

    

plt.plot(y_train[:,1], y_train_pred[:,1], 'x');


    

In [18]:

    

plt.plot(y_train[:,2], y_train_pred[:,2], 'x');


    

In [19]:

    

y_test_pred = model.predict(x_test)


    

In [20]:

    

plt.plot(y_test[:,0], y_test_pred[:,0], 'x');


    

In [21]:

    

plt.plot(y_test[:,1], y_test_pred[:,1], 'x');


    

In [22]:

    

plt.plot(y_test[:,2], y_test_pred[:,2], 'x');