-
Notifications
You must be signed in to change notification settings - Fork 0
/
data.py
174 lines (151 loc) · 6.79 KB
/
data.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
import numpy as np
import pandas as pd
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow.keras.layers import LSTM, Dropout, Dense
from pandas_datareader import data as pdr
from sklearn.metrics import mean_squared_error
from statsmodels.tsa.seasonal import seasonal_decompose
from statsmodels.tsa.vector_ar.var_model import VAR
from statsmodels.tsa.statespace.varmax import VARMAX
from statsmodels.tsa.stattools import adfuller, grangercausalitytests, pacf
from sklearn.preprocessing import StandardScaler
import matplotlib.pyplot as plt
# import yfinance as yfin
# yfin.pdr_override()
# initializing Parameters
start = "1997-01-01"
end = "2021-01-01"
stock_symbols = ["MSFT", "AMD", "AMZN", "GE", "AAPL", "PFE", "KO", "WMT", "COST", "T", "CMCSA", "PEP",
"KO", "UL", "BAC", "JPM", "GS", "MS", "NOC", "LMT", "BA", "F", "LUV", "SLB", "NEE", "MCD",
"SBAC", "AMT", "JNJ"] + ["^GSPC", "^DJI", "^IXIC", "^NYA", "^RUT", "^N100"] + ["JPY=X"]
commodity_symbols = ["^IRX", "^FVX", "^TNX"] + ["FDCAX", "GMCFX", "FRESX", "FRRSX", "FSLBX", "VDIGX", "SGGDX", "OPGSX"] \
+ ["CL", "HE", "B", "NBP", "GL"]
# ["GC=F", "SI=F", "HG=F", "CL=F", "NG=F"]
interval = 'd'
n_periods = 176 # 176 days prediction into future
n_future = 1 # number of future stock prices to predict while training
n_lags = 14 # number of past stock price to consider while training
forecasted = []
# causality_data = []
master_data = pdr.get_data_yahoo(stock_symbols + commodity_symbols, start, end, interval=interval)['Adj Close']
master_data.dropna(inplace=True)
for i in range(len(stock_symbols)):
data = master_data.loc[:, [stock_symbols[i]] + commodity_symbols]
print(data.shape)
# symbols = [stock_symbol] + commodity_symbols
# Getting the data
# data = pdr.get_data_yahoo(symbols, start, end, interval=interval)['Adj Close']
# print(data.shape)
# fill NaN values with back fill
# data.fillna(method='bfill', inplace=True)
# data.dropna(inplace=True)
# normalizing the data
scaler = StandardScaler()
scaler = scaler.fit(data)
data_scaled = scaler.transform(data)
trainX = []
trainY = []
for index in range(n_lags, len(data_scaled) - n_future + 1):
trainX.append(data_scaled[index - n_lags: index, :data_scaled.shape[1]])
trainY.append(data_scaled[index + n_future - 1: index + n_future, 0])
trainX, trainY = np.array(trainX), np.array(trainY)
# print(trainX.shape, trainY.shape)
# model = Sequential()
# model.add(LSTM(64, activation='relu', input_shape=(trainX.shape[1], trainX.shape[2]), return_sequences=True))
# model.add(LSTM(32, activation='relu', return_sequences=False))
# model.add(Dropout(0.2))
# model.add(Dense(trainY.shape[1]))
# model.compile(optimizer='adam', loss='mse')
# # model.summary()
#
# checkpoint = ModelCheckpoint(filepath='model_{0}'.format(stock_symbol), monitor='val_loss',
# verbose = 2, save_best_only = True, mode ='min')
#
# history = model.fit(trainX, trainY, epochs=10, batch_size=16, validation_split=0.1, verbose=2,
# callbacks=[checkpoint])
#
# plt.plot(history.history['loss'])
# plt.plot(history.history['val_loss'])
# plt.title('model loss')
# plt.ylabel('loss')
# plt.xlabel('epoch')
# plt.show()
best_model = tf.keras.models.load_model('model_{0}'.format(stock_symbols[i]))
forecasted_prices_scaled = best_model.predict(trainX[-n_periods:])
forecast_copies = np.repeat(forecasted_prices_scaled, data_scaled.shape[1], axis=-1)
forecasted_prices = scaler.inverse_transform(forecast_copies)[:, 0]
# print(forecasted_prices.shape)
forecasted.append(list(forecasted_prices))
# print(data[-n_periods:].index)
# print(forecasted_prices.shape, data.iloc[-n_periods:, 0].shape)
# print("MSE for {0}: ".format(stock_symbol), mean_squared_error(data.iloc[-n_periods:, 0], forecasted_prices))
# plt.title(stock_symbol)
# plt.xlabel('Datetime')
# plt.ylabel('Adjusted Closing Prices')
# plt.plot(data[-n_periods:].index, data.iloc[-n_periods:, 0])
# plt.plot(data[-n_periods:].index, forecasted_prices)
# plt.legend(['Actual', 'Forecasted'])
# plt.show()
# # Display
# plt.figure(figsize=(20, 10))
# plt.xlabel('Datetime')
# plt.ylabel('Daily Prices ($)')
# plt.title('Closing Prices from {} to {}'.format(start, end))
# plt.plot(data)
# plt.legend([stock_symbol] + commodity_symbols)
# plt.show()
# print(data.isnull().sum())
# print(data.corr())
# # if p-value of this test is >0.05 then data is non-stationary which has to be made stationary for ARIMA to work
# opt_diff = float('-inf')
# for symbol in symbols:
# diff = 0
# while True:
# if diff == 0:
# adfuller_result = adfuller(data[symbol])
# else:
# adfuller_result = adfuller(data[symbol].diff(periods=diff)[diff:])
#
# pvalue = adfuller_result[1]
# if pvalue < .05:
# break
# else:
# diff += 1
# opt_diff = max(diff, opt_diff)
#
# causality test
# causality_data.append([])
# for index in range(len(symbols) - 1):
# # print("Causality for {0} and {1}".format(stock_symbol, symbols[index + 1]))
# try:
# causality = grangercausalitytests(data.iloc[:, [0, index + 1]], maxlag=15, verbose=False)
# ssr_ftest_pvalue = causality[n_lags][0]['ssr_ftest'][1]
# causality_data[i].append(ssr_ftest_pvalue)
# except ValueError:
# causality_data[i].append('NaN')
# continue
#
# # split into train and test
# train_data = data.iloc[:-30, :]
# test_data = data.iloc[-30:]
#
# # finding optimal lags
# model = VAR(train_data.diff(periods=opt_diff)[opt_diff:])
# sorted_model = model.select_order(maxlags=10)
# p = np.argmin(sorted_model)
#
# # train the model
# var_model = VARMAX(train_data, order=(1, 0), enforce_stationarity=True)
# fitted_model = var_model.fit(disp=False)
#
# predict = fitted_model.get_prediction(start=len(train_data), end=len(train_data) + len(test_data) - 1)
# predictions = predict.predicted_mean
#
# plt.plot(test_data.index, test_data[stock_symbol])
# plt.plot(test_data.index, predictions[stock_symbol])
# plt.show()
forecasted = np.array(forecasted).T
df = pd.DataFrame(forecasted, index=master_data[-n_periods:].index, columns=stock_symbols)
df.to_csv('./predictions.csv')