

In [1]:

    
%matplotlib inline  
import matplotlib.pyplot as plt
import numpy as np
import theano
import theano.tensor as T

Getting the data



In [38]:

    
import pandas.io.data as web
import datetime
start = datetime.datetime(2013, 1, 1)
end = datetime.datetime(2013, 1, 27)
f = web.DataReader("F", 'yahoo', start, end)



In [39]:

    
f

Explore data



In [3]:

    
f.Close.plot()









    Out[3]:





<matplotlib.axes._subplots.AxesSubplot at 0x113a174e0>



In [4]:

    
f.Volume.plot()









    Out[4]:





<matplotlib.axes._subplots.AxesSubplot at 0x1144f7f60>



In [5]:

    
f.plot(kind='scatter', x='Close', y='Open')









    Out[5]:





<matplotlib.axes._subplots.AxesSubplot at 0x1146024e0>



In [6]:

    
from pandas.tools.plotting import scatter_matrix
scatter_matrix(f, alpha=0.2, figsize=(6, 6), diagonal='kde');



In [7]:

    
f.describe()









    Out[7]:






  
    
      
      Open
      High
      Low
      Close
      Volume
      Adj Close
    
  
  
    
      count
      17.000000
      17.000000
      17.000000
      17.000000
      1.700000e+01
      17.000000
    
    
      mean
      13.779412
      13.900000
      13.644706
      13.808824
      5.665848e+07
      12.778609
    
    
      std
      0.334449
      0.315099
      0.351196
      0.345866
      2.118857e+07
      0.320063
    
    
      min
      13.230000
      13.280000
      13.000000
      13.200000
      3.546790e+07
      12.215207
    
    
      25%
      13.510000
      13.610000
      13.350000
      13.470000
      4.429930e+07
      12.465064
    
    
      50%
      13.830000
      13.980000
      13.790000
      13.870000
      5.145340e+07
      12.835221
    
    
      75%
      14.040000
      14.130000
      13.900000
      14.110000
      5.812220e+07
      13.057316
    
    
      max
      14.270000
      14.300000
      14.070000
      14.300000
      1.212847e+08
      13.233141

Data Cleaning



In [8]:

    
Xtr = f[['Open','Low','High']].values.astype(np.float32)
Xtr









    Out[8]:





array([[ 13.22999954,  13.        ,  13.27999973],
       [ 13.23999977,  13.05000019,  13.69999981],
       [ 13.51000023,  13.35000038,  13.60999966],
       [ 13.52000046,  13.35000038,  13.57999992],
       [ 13.38000011,  13.19999981,  13.43000031],
       [ 13.39999962,  13.39000034,  13.60000038],
       [ 13.77999973,  13.68000031,  13.93999958],
       [ 14.        ,  13.89999962,  14.06999969],
       [ 14.03999996,  13.85999966,  14.07999992],
       [ 13.90999985,  13.89000034,  14.30000019],
       [ 14.15999985,  14.01000023,  14.25      ],
       [ 14.27000046,  14.06999969,  14.30000019],
       [ 14.10000038,  13.97000027,  14.13000011],
       [ 14.06000042,  14.        ,  14.18999958],
       [ 14.        ,  13.78999996,  14.02000046],
       [ 13.81999969,  13.81000042,  13.97999954],
       [ 13.82999992,  13.64000034,  13.84000015]], dtype=float32)



In [9]:

    
Ytr = f.Close.values.astype(np.float32)
Ytr









    Out[9]:





array([ 13.19999981,  13.46000004,  13.56999969,  13.43000031,
        13.35000038,  13.47000027,  13.82999992,  14.        ,
        13.98999977,  14.30000019,  14.22000027,  14.22000027,
        14.10999966,  14.17000008,  13.88000011,  13.86999989,  13.68000031], dtype=float32)

Modeling



In [10]:

    
rng = np.random



In [11]:

    
def define_fit_model(dataset_x, dataset_y, learning_rate = 1e-4, training_steps = 10000):
    N = len(dataset_x[0])
    x = T.matrix("x")
    y = T.vector("y")
    w = theano.shared(rng.randn(N), name="w")
    b = theano.shared(0., name="b")
    Hx = T.dot(x,w) + b
    cost = T.mean(T.sqr(Hx - y))
    gw, gb = T.grad(cost=cost, wrt=[w, b])
    train = theano.function(inputs=[x,y],outputs=[Hx,cost], updates=((w, w - learning_rate * gw), (b, b - learning_rate * gb)))
    predict = theano.function(inputs=[x], outputs=Hx)
    for i in range(training_steps):
        pred, err = train(dataset_x, dataset_y)
    return predict, w, b



In [12]:

    
predict, w, b = define_fit_model(Xtr,Ytr)



In [13]:

    
"Model that was fit from data:\n {}, {}".format( w.get_value(), b.get_value() )









    Out[13]:





'Model that was fit from data:\n [-1.0897295   1.04515748  1.04634064], 0.019895761707419128'

Visualize, Summarize, Apply

predict



In [23]:

    
Ypredict = predict(Xtr)



In [24]:

    
plt.plot(Ytr, label='real')
plt.plot(Ypredict, label='predict')
plt.legend();

testing



In [34]:

    
start = datetime.datetime(2015, 5, 1)
end = datetime.datetime(2015, 5, 30)
f = web.DataReader("F", 'yahoo', start, end)



In [35]:

    
Xts = f[['Open','Low','High']].values.astype(np.float32)
Yts = f.Close.values.astype(np.float32)



In [36]:

    
Ytest_predict = predict(Xts)



In [37]:

    
plt.plot(Yts, label='test')
plt.plot(Ytest_predict, label='predict')
plt.legend();



In [ ]:

	Open	High	Low	Close	Volume	Adj Close
Date
2013-01-02	13.23	13.28	13.00	13.20	75274700	12.215207
2013-01-03	13.24	13.70	13.05	13.46	121284700	12.455810
2013-01-04	13.51	13.61	13.35	13.57	54669900	12.557603
2013-01-07	13.52	13.58	13.35	13.43	43482400	12.428048
2013-01-08	13.38	13.43	13.20	13.35	46336200	12.354017
2013-01-09	13.40	13.60	13.39	13.47	36973900	12.465064
2013-01-10	13.78	13.94	13.68	13.83	84884300	12.798206
2013-01-11	14.00	14.07	13.90	14.00	67641200	12.955523
2013-01-14	14.04	14.08	13.86	13.99	46224800	12.946268
2013-01-15	13.91	14.30	13.89	14.30	55425000	13.233141
2013-01-16	14.16	14.25	14.01	14.22	51453400	13.159110
2013-01-17	14.27	14.30	14.07	14.22	44299300	13.159110
2013-01-18	14.10	14.13	13.97	14.11	45716100	13.057316
2013-01-22	14.06	14.19	14.00	14.17	35467900	13.112840
2013-01-23	14.00	14.02	13.79	13.88	58122200	12.844475
2013-01-24	13.82	13.98	13.81	13.87	42532700	12.835221
2013-01-25	13.83	13.84	13.64	13.68	53405400	12.659397

	Open	High	Low	Close	Volume	Adj Close
count	17.000000	17.000000	17.000000	17.000000	1.700000e+01	17.000000
mean	13.779412	13.900000	13.644706	13.808824	5.665848e+07	12.778609
std	0.334449	0.315099	0.351196	0.345866	2.118857e+07	0.320063
min	13.230000	13.280000	13.000000	13.200000	3.546790e+07	12.215207
25%	13.510000	13.610000	13.350000	13.470000	4.429930e+07	12.465064
50%	13.830000	13.980000	13.790000	13.870000	5.145340e+07	12.835221
75%	14.040000	14.130000	13.900000	14.110000	5.812220e+07	13.057316
max	14.270000	14.300000	14.070000	14.300000	1.212847e+08	13.233141