-
Notifications
You must be signed in to change notification settings - Fork 0
/
recursivemlp.py
353 lines (327 loc) · 18.1 KB
/
recursivemlp.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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
import pandas as pd
import numpy as np
import keras
import os
import time
import tensorflow
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Dropout
from tensorflow.keras.callbacks import EarlyStopping, LearningRateScheduler
from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import StandardScaler
from sklearn.feature_selection import mutual_info_regression
from sklearn.model_selection import ParameterGrid
from numpy.random import seed
from tqdm import tqdm
from datetime import datetime, timedelta
import json
import json
from collections import Counter
import warnings
from os import path
from joblib import Parallel, delayed
import multiprocessing
seed(123)
tensorflow.random.set_seed(123)
class MetricsValidation(keras.callbacks.Callback):
def __init__(self, model, validation_data, selectedfeatures, lastprices, verbose):
super().__init__()
self.model = model
self.validation_data = validation_data
self.selectedfeatures = selectedfeatures
self.lastprices = lastprices
self.verbose = verbose
# def on_train_begin(self, logs={}):
# self.iteratedmae = 0
def on_epoch_end(self, epoch, logs={}):
validationdays = int(self.validation_data[0].shape[0] / 24)
y_hatlist = []
for validationday in range(validationdays, 0, -1):
yhat_dailylist = []
if validationday > 1:
validationdaily = self.validation_data[0][-24 * validationday:-24 * (validationday - 1)]
else:
validationdaily = self.validation_data[0][-24 * validationday:]
lastprices = self.lastprices[-24 * (validationday + 1):-24 * validationday].flatten().tolist()
lastprices.reverse()
for hour in range(24):
pricedict = {}
for lag in range(1, 25):
pricedict['price difference t_' + str(lag)] = lastprices[lag - 1]
validation_x = validationdaily[hour].flatten().reshape(1, -1)
if len(yhat_dailylist) > 0:
for feature in self.selectedfeatures:
if feature in pricedict.keys():
featureindex = self.selectedfeatures.index(feature)
validation_x[0, featureindex] = pricedict[feature]
yhat = float(self.model.predict(validation_x))
yhat_dailylist.append(yhat)
lastprices.insert(0, yhat)
del (lastprices[-1])
y_hatlist.append(np.asarray(yhat_dailylist))
y_pred = np.asarray(y_hatlist).flatten()
y_true = np.asarray(self.validation_data[1][-24 * validationdays:]).flatten()
m = tensorflow.keras.metrics.MeanAbsoluteError()
iteratedmae = m(y_pred, y_true)
logs['iteratedmae'] = iteratedmae
if self.verbose:
print('-val-iterated: %f' % np.round(iteratedmae, decimals=4))
def miselection(df, feature_number, allfeatures):
y = df['price difference']
X = df[allfeatures]
mi = mutual_info_regression(X[:-24], y[:-24], n_neighbors=5)
mi /= np.max(mi)
dictionary = dict(zip(allfeatures, mi))
sorted_dict = sorted(dictionary.items(), key=lambda x: x[1], reverse=True)
sorted_features = [sorted_dict[i][0] for i in range(len(sorted_dict))]
selected_features = sorted_features[:feature_number]
return selected_features
def collectingdata(inputdf, predictiondate, lookbackyears, trainingdayscurrent, trainingdaysprevious, testdays):
datebegincurrent = predictiondate + timedelta(days=-(trainingdayscurrent))
datacurrent = inputdf.loc[(datebegincurrent.strftime('%Y-%m-%d') + ' 00'):(
predictiondate.strftime('%Y-%m-%d') + ' 23'), :]
timestamps = datacurrent.index.to_list()[-int(24 * testdays):]
if lookbackyears > 0:
combinedprevious = None
for year in range(lookbackyears, 0, -1):
datebeginprevious = predictiondate + timedelta(days=-(trainingdaysprevious + 365 * year))
dateendprevious = datebeginprevious + timedelta(days=2 * trainingdaysprevious - 1)
dataprevious = inputdf.loc[(datebeginprevious.strftime('%Y-%m-%d') + ' 00'):(
dateendprevious.strftime('%Y-%m-%d') + ' 23'), :]
if combinedprevious is None:
combinedprevious = dataprevious
else:
combinedprevious = pd.concat([combinedprevious, dataprevious], ignore_index=True)
datacombined = pd.concat([combinedprevious, datacurrent], ignore_index=True).dropna()
else:
datacombined = datacurrent.dropna()
return timestamps, datacombined
def modelbuilding(hiddenlayersizes, featurenum, activation, dropout):
model = Sequential()
# if inputlayer dropout
# model.add(Dropout(0.1,inputshape=()))
# model.add(Dense(hiddenlayersizes[0],init=...))
model.add(
Dense(hiddenlayersizes[0], input_dim=featurenum, kernel_initializer='normal', activation=activation))
if dropout > 0:
model.add(Dropout(dropout, seed=123))
for layersize in hiddenlayersizes[1:]:
if dropout > 0:
model.add(Dense(layersize, activation=activation))
model.add(Dropout(dropout, seed=123))
else:
model.add(Dense(layersize, activation=activation))
model.add(Dense(1, activation='linear'))
return model
def trainandpredict(totaldata, scaler, selectedfeatures, validationdays, testdays, hiddenlayersizes, dropout,
activation, epochs, batch_size, patience, verbose):
if scaler == 'MinMaxScaler':
scaler_x = MinMaxScaler(feature_range=(0, 1))
scaler_y = MinMaxScaler(feature_range=(0, 1))
else:
scaler_x = StandardScaler()
scaler_y = StandardScaler()
scaled_x = scaler_x.fit_transform(totaldata[:-int(testdays * 24), 1:])
scaled_y = scaler_y.fit_transform(totaldata[:-int(testdays * 24), 0].reshape(-1, 1))
train_x = scaled_x[:-int(validationdays * 24), :]
train_y = scaled_y[:-int(validationdays * 24), :]
validation_x = scaled_x[-int(validationdays * 24):, :]
validation_y = scaled_y[-int(validationdays * 24):, :]
test = totaldata[-int(testdays * 24):, :]
featurenum = len(selectedfeatures)
model = modelbuilding(hiddenlayersizes=hiddenlayersizes, featurenum=featurenum, activation=activation,
dropout=dropout)
# print(model.summary())
model.compile(loss='mae', optimizer=tensorflow.keras.optimizers.Adam(lr=0.001), metrics=['mae'])
lastprices = scaled_y[-24 * (validationdays + 1):]
# metricsvalidation = MetricsValidation(model=model, validation_data=(validation_x, validation_y),
# selectedfeatures=selectedfeatures, lastprices=lastprices, verbose=verbose)
es = EarlyStopping(monitor='val_loss', mode='min', verbose=verbose, patience=patience, restore_best_weights=True)
model.fit(train_x, train_y, epochs=epochs, batch_size=batch_size, verbose=verbose,
validation_data=(validation_x, validation_y), callbacks=[es])
y_hatlist = []
for testday in range(testdays, 0, -1):
yhat_dailylist = []
if testday > 1:
testdaily = test[-int(24 * testday):-int(24 * (testday - 1)), :]
else:
testdaily = test[-int(24 * testday):, :]
lastprices = totaldata[-int(23 + testday * 24):-int(testday * 24), 0].tolist()
lastprices.reverse()
for hour in range(24):
pricedict = {}
# the last lagged residul (price difference) are changed for each hour. The predicted values is used instead
# of real values. If a new value is predicted, it is added to the lastprices list
for lag in range(1, 24):
pricedict['price difference t_' + str(lag)] = lastprices[lag - 1]
test_x = testdaily[hour, 1:].reshape(1, -1)
if len(yhat_dailylist) > 0:
for feature in selectedfeatures:
if feature in pricedict.keys():
featureindex = selectedfeatures.index(feature)
test_x[0, featureindex] = pricedict[feature]
scaled_x = scaler_x.transform(test_x)
yhat = model.predict(scaled_x)
yhat = float(scaler_y.inverse_transform(yhat))
yhat_dailylist.append(yhat)
lastprices.insert(0, yhat)
del (lastprices[-1])
y_hatlist.append(np.asarray(yhat_dailylist))
y = test[:, 0].tolist()
return np.asarray(y_hatlist).flatten(), np.asarray(y)
def run(daylist):
yhatlist = []
ylist = []
timestamplist = []
sectionbegin = datetime.strptime(predictionbegin, '%Y-%m-%d') + timedelta(days=daylist[0])
sectionbegin = sectionbegin.strftime('%Y-%m-%d')
if testdays == 1:
for i in tqdm(daylist):
predictiondate = datetime.strptime(predictionbegin, '%Y-%m-%d') + timedelta(days=i)
datum = predictiondate.strftime('%Y-%m-%d')
timestamps, datacombined = collectingdata(inputdftmp, predictiondate, params['lookbackyears'],
params['trainingdayscurrent'], params['trainingdaysprevious'],
testdays)
if featureselection == 'mi':
selectedfeatures = miselection(datacombined, featurenumber, features)
datacombined = datacombined[['price difference'] + selectedfeatures].values
yhat, y = trainandpredict(totaldata=datacombined, scaler=scaler,
selectedfeatures=selectedfeatures,
validationdays=validationdays, testdays=testdays,
hiddenlayersizes=hiddenlayersizes, dropout=dropout,
activation=activation, epochs=epochs, batch_size=batchsize,
patience=patience, verbose=verbose)
timestamplist.append(timestamps)
yhatlist.append(yhat)
ylist.append(y)
if featureselection == 'None':
selectedfeatures = features
datacombined = datacombined[['price difference'] + selectedfeatures].values
yhat, y = trainandpredict(totaldata=datacombined, scaler=scaler,
selectedfeatures=selectedfeatures,
validationdays=validationdays, testdays=testdays,
hiddenlayersizes=hiddenlayersizes, dropout=dropout,
activation=activation, epochs=epochs, batch_size=batchsize,
patience=patience, verbose=verbose)
timestamplist.append(timestamps)
yhatlist.append(yhat)
ylist.append(y)
if featureselection == 'mrmrensemble':
# selectedfeaturelist = mrmrensembleselection(datacombined, featurenumber, features)
selectedfeaturelist = selectedfeaturedict[datum]
y_hatensemblelist = []
for selectedfeatures in selectedfeaturelist[:1]:
tmp = datacombined[['price difference'] + selectedfeatures].values
yhat, y = trainandpredict(totaldata=tmp, scaler=scaler, selectedfeatures=selectedfeatures,
validationdays=validationdays, testdays=testdays,
hiddenlayersizes=hiddenlayersizes, dropout=dropout,
activation=activation, epochs=epochs, batch_size=batchsize,
patience=patience, verbose=verbose)
y_hatensemblelist.append(yhat)
yhat = np.mean(y_hatensemblelist, axis=0)
timestamplist.append(timestamps)
yhatlist.append(yhat)
ylist.append(y)
else:
for i in [daydeltalist[j * 7] for j in range(int(len(daydeltalist) / 7))]:
predictiondate = datetime.strptime(predictionbegin, '%Y-%m-%d') + timedelta(days=i + 6)
timestamps, datacombined = collectingdata(inputdftmp, predictiondate, lookbackyears,
trainingdayscurrent, trainingdaysprevious, testdays)
if featureselection == 'mi':
selectedfeatures = miselection(datacombined, featurenumber, features)
datacombined = datacombined[['price difference'] + selectedfeatures].values
yhat, y = trainandpredict(totaldata=datacombined, scaler=scaler,
selectedfeatures=selectedfeatures,
validationdays=validationdays, testdays=testdays,
hiddenlayersizes=hiddenlayersizes, dropout=dropout,
activation=activation, epochs=epochs, batch_size=batchsize,
patience=patience, verbose=verbose)
timestamplist.append(timestamps)
yhatlist.append(yhat)
ylist.append(y)
if featureselection == 'None':
selectedfeatures = features
datacombined = datacombined[['price difference'] + selectedfeatures].values
yhat, y = trainandpredict(totaldata=datacombined, scaler=scaler,
selectedfeatures=selectedfeatures,
validationdays=validationdays, testdays=testdays,
hiddenlayersizes=hiddenlayersizes, dropout=dropout,
activation=activation, epochs=epochs, batch_size=batchsize,
patience=patience, verbose=verbose)
timestamplist.append(timestamps)
yhatlist.append(yhat)
ylist.append(y)
if featureselection == 'mrmrensemble':
# selectedfeaturelist = mrmrensembleselection(datacombined, featurenumber, features)
selectedfeaturelist = selectedfeaturedict['datum']
y_hatensemblelist = []
for selectedfeatures in selectedfeaturelist:
tmp = datacombined[['price difference'] + selectedfeatures].values
yhat, y = trainandpredict(totaldata=tmp, scaler=scaler, selectedfeatures=selectedfeatures,
validationdays=validationdays, testdays=testdays,
hiddenlayersizes=hiddenlayersizes, dropout=dropout,
activation=activation, epochs=epochs, batch_size=batchsize,
patience=patience, verbose=verbose)
y_hatensemblelist.append(yhat)
yhat = np.mean(y_hatensemblelist, axis=0)
timestamplist.append(timestamps)
yhatlist.append(yhat)
ylist.append(y)
predictionlist = np.asarray(yhatlist).flatten()
actuallist = np.asarray(ylist).flatten()
timestamplist = np.asarray(timestamplist).flatten()
error = actuallist - predictionlist
predictiondf = pd.DataFrame(
{'timestamp': timestamplist, 'prediction': predictionlist, 'actual': actuallist, 'error': error})
filename = sectionbegin + '_' + remarks
filepath = path.join('Prediction/RecursiveMLP', filename + '.' + 'csv')
predictiondf.to_csv(filepath, sep=';', index=False)
if __name__ == '__main__':
with open('Config/config_recursivemlp.json', 'r') as fp:
parameterdict = json.load(fp)
inputfilepath = parameterdict['inputfilepath']
inputdf = pd.read_csv(inputfilepath, sep=';', index_col=['UTC Timestamp'])
predictionbegin = parameterdict['predictionbegin']
daystopredict = parameterdict['daystopredict']
validationdays = parameterdict['validationdays']
testdays = parameterdict['testdays']
runtimes = parameterdict['runtimes']
verbose = parameterdict['verbose']
# parameters for the deep neural networks
hyperparameters = parameterdict['hyperparameters']
params_grid = ParameterGrid(hyperparameters)
# read json file for days to predict
with open('Data/sampledays2018.json', 'r') as fp:
daylist = json.load(fp)
daydeltalist = [
int((datetime.strptime(item, "%Y-%m-%d") - datetime.strptime(predictionbegin,
"%Y-%m-%d")).total_seconds() / 86400)
for item in daylist]
# with open('bench_all24_selectedfeature_mrmrensemble.json', 'r') as fp:
# selectedfeaturedict = json.load(fp)
selectedfeaturedict = {}
for enum, params in enumerate(tqdm(params_grid)):
lags = params['lags']
lookbackyears = params['lookbackyears']
trainingdayscurrent = params['trainingdayscurrent']
trainingdaysprevious = params['trainingdaysprevious']
featureselection = params['featureselection']
scaler = params['scaler']
hiddenlayersizes = params['hiddenlayersizes']
dropout = params['dropout']
activation = params['activation']
patience = params['patience']
epochs = params['epochs']
batchsize = params['batchsize']
# add the features with the p consecutive lagged variable
features = parameterdict['features'] + ['price difference t_' + str(i) for i in range(1, lags + 1)]
featurenumber = parameterdict['featurenumber']
allcolumns = ['price difference'] + features
inputdftmp = inputdf[allcolumns]
remarks = 'Iterative' + 'LB' + str(lookbackyears) + 'VD' + str(validationdays) + 'Lag' + str(
lags) + 'hiddenL' + str(len(hiddenlayersizes)) + 'S' + str(
hiddenlayersizes[0]) + featureselection
#Parallel(n_jobs=-1)(delayed(run)(item) for item in
# [list(range(0, 90)), list(range(90, 180)), list(range(180, 270)), list(range(270, 365))])
run(list(range(365)))