Time Series Analysis - Deep Learning for Time Series Forecasting
Deep Learning for Time Series Forecasting
import tensorflow.keras
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
y = mx + b + noise
m = 2
b = 3
x = np.linspace(0, 50, 100)
np.random.seed(101)
noise = np.random.normal(loc=0, scale=4, size=len(x))
y = 2*x + b + noise
plt.plot(x, y, '*')
[<matplotlib.lines.Line2D at 0x1a4db22fc18>]
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
model = Sequential()
model.add(Dense(4, input_dim=1,activation='relu'))
model.add(Dense(4, activation ='relu'))
model.add(Dense(1, activation ='relu'))
model.compile(loss = 'mse')
model.summary()
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
dense_3 (Dense) (None, 4) 8
_________________________________________________________________
dense_4 (Dense) (None, 4) 20
_________________________________________________________________
dense_5 (Dense) (None, 1) 5
=================================================================
Total params: 33
Trainable params: 33
Non-trainable params: 0
_________________________________________________________________
model.fit(x,y,epochs=200)
Epoch 1/200
4/4 [==============================] - 0s 1ms/step - loss: 1835.2828
Epoch 2/200
4/4 [==============================] - 0s 1ms/step - loss: 1760.2566
Epoch 3/200
4/4 [==============================] - 0s 1ms/step - loss: 1708.3179
Epoch 4/200
4/4 [==============================] - 0s 1ms/step - loss: 1661.5303
Epoch 5/200
4/4 [==============================] - 0s 1ms/step - loss: 1622.3416
Epoch 6/200
4/4 [==============================] - 0s 2ms/step - loss: 1580.1083
Epoch 7/200
4/4 [==============================] - 0s 1ms/step - loss: 1546.7878
Epoch 8/200
4/4 [==============================] - 0s 1ms/step - loss: 1506.9521
Epoch 9/200
4/4 [==============================] - 0s 1ms/step - loss: 1470.2085
Epoch 10/200
4/4 [==============================] - 0s 1ms/step - loss: 1427.7432
Epoch 11/200
4/4 [==============================] - 0s 1ms/step - loss: 1386.2589
Epoch 12/200
4/4 [==============================] - 0s 1ms/step - loss: 1352.8347
Epoch 13/200
4/4 [==============================] - 0s 1ms/step - loss: 1323.4415
Epoch 14/200
4/4 [==============================] - 0s 1ms/step - loss: 1288.4734
Epoch 15/200
4/4 [==============================] - 0s 1ms/step - loss: 1254.3790
Epoch 16/200
4/4 [==============================] - 0s 1ms/step - loss: 1213.1908
Epoch 17/200
4/4 [==============================] - 0s 1ms/step - loss: 1177.0500
Epoch 18/200
4/4 [==============================] - 0s 1ms/step - loss: 1143.3599
Epoch 19/200
4/4 [==============================] - 0s 2ms/step - loss: 1105.8571
Epoch 20/200
4/4 [==============================] - 0s 1ms/step - loss: 1072.1377
Epoch 21/200
4/4 [==============================] - 0s 1ms/step - loss: 1034.3806
Epoch 22/200
4/4 [==============================] - 0s 1ms/step - loss: 997.8810
Epoch 23/200
4/4 [==============================] - 0s 1ms/step - loss: 962.5549
Epoch 24/200
4/4 [==============================] - 0s 1ms/step - loss: 929.9268
Epoch 25/200
4/4 [==============================] - 0s 1ms/step - loss: 895.7763
Epoch 26/200
4/4 [==============================] - 0s 2ms/step - loss: 861.8133
Epoch 27/200
4/4 [==============================] - 0s 1ms/step - loss: 825.4382
Epoch 28/200
4/4 [==============================] - 0s 1ms/step - loss: 788.8192
Epoch 29/200
4/4 [==============================] - 0s 1ms/step - loss: 754.3546
Epoch 30/200
4/4 [==============================] - 0s 1ms/step - loss: 720.8946
Epoch 31/200
4/4 [==============================] - 0s 2ms/step - loss: 694.4216
Epoch 32/200
4/4 [==============================] - 0s 1ms/step - loss: 665.4741
Epoch 33/200
4/4 [==============================] - 0s 1ms/step - loss: 629.9188
Epoch 34/200
4/4 [==============================] - 0s 2ms/step - loss: 598.7121
Epoch 35/200
4/4 [==============================] - 0s 1ms/step - loss: 567.5757
Epoch 36/200
4/4 [==============================] - 0s 1ms/step - loss: 537.9247
Epoch 37/200
4/4 [==============================] - 0s 1ms/step - loss: 510.0300
Epoch 38/200
4/4 [==============================] - 0s 1ms/step - loss: 480.5556
Epoch 39/200
4/4 [==============================] - 0s 1ms/step - loss: 455.6969
Epoch 40/200
4/4 [==============================] - 0s 1ms/step - loss: 429.8373
Epoch 41/200
4/4 [==============================] - 0s 1ms/step - loss: 402.3074
Epoch 42/200
4/4 [==============================] - 0s 1ms/step - loss: 376.8388
Epoch 43/200
4/4 [==============================] - 0s 1ms/step - loss: 351.1939
Epoch 44/200
4/4 [==============================] - 0s 1ms/step - loss: 329.2676
Epoch 45/200
4/4 [==============================] - 0s 1ms/step - loss: 305.4028
Epoch 46/200
4/4 [==============================] - 0s 1ms/step - loss: 284.1473
Epoch 47/200
4/4 [==============================] - 0s 1ms/step - loss: 263.4900
Epoch 48/200
4/4 [==============================] - 0s 2ms/step - loss: 243.5607
Epoch 49/200
4/4 [==============================] - 0s 1ms/step - loss: 219.7183
Epoch 50/200
4/4 [==============================] - 0s 1ms/step - loss: 202.2714
Epoch 51/200
4/4 [==============================] - 0s 1ms/step - loss: 181.6279
Epoch 52/200
4/4 [==============================] - 0s 1ms/step - loss: 164.6638
Epoch 53/200
4/4 [==============================] - 0s 1ms/step - loss: 149.9858
Epoch 54/200
4/4 [==============================] - 0s 1ms/step - loss: 135.7605
Epoch 55/200
4/4 [==============================] - 0s 1ms/step - loss: 120.9801
Epoch 56/200
4/4 [==============================] - 0s 1ms/step - loss: 106.5698
Epoch 57/200
4/4 [==============================] - 0s 1ms/step - loss: 94.0438
Epoch 58/200
4/4 [==============================] - 0s 1ms/step - loss: 81.3139
Epoch 59/200
4/4 [==============================] - 0s 1ms/step - loss: 72.4593
Epoch 60/200
4/4 [==============================] - 0s 1ms/step - loss: 63.2077
Epoch 61/200
4/4 [==============================] - 0s 1ms/step - loss: 54.8502
Epoch 62/200
4/4 [==============================] - 0s 1ms/step - loss: 48.2167
Epoch 63/200
4/4 [==============================] - 0s 1ms/step - loss: 42.0426
Epoch 64/200
4/4 [==============================] - 0s 1ms/step - loss: 36.3565
Epoch 65/200
4/4 [==============================] - 0s 2ms/step - loss: 31.2589
Epoch 66/200
4/4 [==============================] - 0s 1ms/step - loss: 26.9042
Epoch 67/200
4/4 [==============================] - 0s 1ms/step - loss: 24.5516
Epoch 68/200
4/4 [==============================] - 0s 1ms/step - loss: 23.0759
Epoch 69/200
4/4 [==============================] - 0s 1ms/step - loss: 21.8395
Epoch 70/200
4/4 [==============================] - 0s 1ms/step - loss: 20.5880
Epoch 71/200
4/4 [==============================] - 0s 1ms/step - loss: 19.7270
Epoch 72/200
4/4 [==============================] - 0s 1ms/step - loss: 19.4509
Epoch 73/200
4/4 [==============================] - 0s 1ms/step - loss: 19.5544
Epoch 74/200
4/4 [==============================] - 0s 1ms/step - loss: 19.4520
Epoch 75/200
4/4 [==============================] - 0s 1ms/step - loss: 19.4514
Epoch 76/200
4/4 [==============================] - 0s 1ms/step - loss: 19.3336
Epoch 77/200
4/4 [==============================] - 0s 1ms/step - loss: 19.2367
Epoch 78/200
4/4 [==============================] - 0s 1ms/step - loss: 19.4528
Epoch 79/200
4/4 [==============================] - 0s 1ms/step - loss: 19.1718
Epoch 80/200
4/4 [==============================] - 0s 1ms/step - loss: 19.2670
Epoch 81/200
4/4 [==============================] - 0s 1ms/step - loss: 19.4927
Epoch 82/200
4/4 [==============================] - 0s 1ms/step - loss: 19.2007
Epoch 83/200
4/4 [==============================] - 0s 1ms/step - loss: 19.4057
Epoch 84/200
4/4 [==============================] - 0s 1ms/step - loss: 19.1590
Epoch 85/200
4/4 [==============================] - 0s 1ms/step - loss: 19.1011
Epoch 86/200
4/4 [==============================] - 0s 1ms/step - loss: 19.2685
Epoch 87/200
4/4 [==============================] - 0s 1ms/step - loss: 19.0871
Epoch 88/200
4/4 [==============================] - 0s 1ms/step - loss: 19.1214
Epoch 89/200
4/4 [==============================] - 0s 1ms/step - loss: 19.1207
Epoch 90/200
4/4 [==============================] - 0s 2ms/step - loss: 19.2000
Epoch 91/200
4/4 [==============================] - 0s 2ms/step - loss: 19.0114
Epoch 92/200
4/4 [==============================] - 0s 1ms/step - loss: 19.0037
Epoch 93/200
4/4 [==============================] - 0s 1ms/step - loss: 19.4451
Epoch 94/200
4/4 [==============================] - 0s 2ms/step - loss: 19.1021
Epoch 95/200
4/4 [==============================] - 0s 1ms/step - loss: 19.0582
Epoch 96/200
4/4 [==============================] - 0s 2ms/step - loss: 19.0108
Epoch 97/200
4/4 [==============================] - 0s 1ms/step - loss: 19.0246
Epoch 98/200
4/4 [==============================] - 0s 2ms/step - loss: 18.9366
Epoch 99/200
4/4 [==============================] - 0s 1ms/step - loss: 18.9903
Epoch 100/200
4/4 [==============================] - 0s 1ms/step - loss: 18.9929
Epoch 101/200
4/4 [==============================] - 0s 1ms/step - loss: 18.9058
Epoch 102/200
4/4 [==============================] - 0s 1ms/step - loss: 18.9509
Epoch 103/200
4/4 [==============================] - 0s 1ms/step - loss: 19.0329
Epoch 104/200
4/4 [==============================] - 0s 1ms/step - loss: 18.9732
Epoch 105/200
4/4 [==============================] - 0s 1ms/step - loss: 18.7965
Epoch 106/200
4/4 [==============================] - 0s 1ms/step - loss: 18.8212
Epoch 107/200
4/4 [==============================] - 0s 1ms/step - loss: 18.7402
Epoch 108/200
4/4 [==============================] - 0s 1ms/step - loss: 18.8385
Epoch 109/200
4/4 [==============================] - 0s 1ms/step - loss: 18.8170
Epoch 110/200
4/4 [==============================] - 0s 1ms/step - loss: 18.8779
Epoch 111/200
4/4 [==============================] - 0s 997us/step - loss: 18.7897
Epoch 112/200
4/4 [==============================] - 0s 1ms/step - loss: 18.9200
Epoch 113/200
4/4 [==============================] - 0s 1ms/step - loss: 18.9472
Epoch 114/200
4/4 [==============================] - 0s 1ms/step - loss: 18.8327
Epoch 115/200
4/4 [==============================] - 0s 1ms/step - loss: 18.7283
Epoch 116/200
4/4 [==============================] - 0s 1ms/step - loss: 18.6641
Epoch 117/200
4/4 [==============================] - 0s 1ms/step - loss: 18.9163
Epoch 118/200
4/4 [==============================] - 0s 1ms/step - loss: 18.5791
Epoch 119/200
4/4 [==============================] - 0s 1ms/step - loss: 18.6373
Epoch 120/200
4/4 [==============================] - 0s 1ms/step - loss: 18.5254
Epoch 121/200
4/4 [==============================] - 0s 1ms/step - loss: 18.5407
Epoch 122/200
4/4 [==============================] - 0s 1ms/step - loss: 18.7938
Epoch 123/200
4/4 [==============================] - 0s 1ms/step - loss: 18.9352
Epoch 124/200
4/4 [==============================] - 0s 1ms/step - loss: 18.7772
Epoch 125/200
4/4 [==============================] - 0s 997us/step - loss: 18.5229
Epoch 126/200
4/4 [==============================] - 0s 1ms/step - loss: 18.5174
Epoch 127/200
4/4 [==============================] - 0s 1ms/step - loss: 18.4520
Epoch 128/200
4/4 [==============================] - 0s 1ms/step - loss: 18.4360
Epoch 129/200
4/4 [==============================] - 0s 1ms/step - loss: 18.4862
Epoch 130/200
4/4 [==============================] - 0s 1ms/step - loss: 18.5956
Epoch 131/200
4/4 [==============================] - 0s 1ms/step - loss: 18.5477
Epoch 132/200
4/4 [==============================] - 0s 1ms/step - loss: 18.5845
Epoch 133/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3726
Epoch 134/200
4/4 [==============================] - 0s 1ms/step - loss: 18.4655
Epoch 135/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3289
Epoch 136/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3460
Epoch 137/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3917
Epoch 138/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3464
Epoch 139/200
4/4 [==============================] - 0s 1ms/step - loss: 18.4719
Epoch 140/200
4/4 [==============================] - 0s 1ms/step - loss: 18.6031
Epoch 141/200
4/4 [==============================] - 0s 1ms/step - loss: 18.4234
Epoch 142/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3097
Epoch 143/200
4/4 [==============================] - 0s 1ms/step - loss: 18.2677
Epoch 144/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3238
Epoch 145/200
4/4 [==============================] - 0s 1ms/step - loss: 18.2502
Epoch 146/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3489
Epoch 147/200
4/4 [==============================] - 0s 1ms/step - loss: 18.2561
Epoch 148/200
4/4 [==============================] - 0s 1ms/step - loss: 18.2409
Epoch 149/200
4/4 [==============================] - 0s 1ms/step - loss: 18.4153
Epoch 150/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1522
Epoch 151/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1873
Epoch 152/200
4/4 [==============================] - 0s 1ms/step - loss: 18.2808
Epoch 153/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1157
Epoch 154/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1804
Epoch 155/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3109
Epoch 156/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3864
Epoch 157/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3333
Epoch 158/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3745
Epoch 159/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3049
Epoch 160/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1015
Epoch 161/200
4/4 [==============================] - 0s 1ms/step - loss: 18.0976
Epoch 162/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1385
Epoch 163/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1505
Epoch 164/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1945
Epoch 165/200
4/4 [==============================] - 0s 1ms/step - loss: 18.2722
Epoch 166/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9870
Epoch 167/200
4/4 [==============================] - ETA: 0s - loss: 18.00 - 0s 1ms/step - loss: 18.0277
Epoch 168/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1463
Epoch 169/200
4/4 [==============================] - 0s 1ms/step - loss: 18.5038
Epoch 170/200
4/4 [==============================] - 0s 2ms/step - loss: 18.1617
Epoch 171/200
4/4 [==============================] - 0s 1ms/step - loss: 18.3358
Epoch 172/200
4/4 [==============================] - 0s 1ms/step - loss: 18.2571
Epoch 173/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9925
Epoch 174/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9974
Epoch 175/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1721
Epoch 176/200
4/4 [==============================] - 0s 1ms/step - loss: 18.0019
Epoch 177/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9564
Epoch 178/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9195
Epoch 179/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9515
Epoch 180/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9477
Epoch 181/200
4/4 [==============================] - 0s 1ms/step - loss: 17.8825
Epoch 182/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9276
Epoch 183/200
4/4 [==============================] - 0s 2ms/step - loss: 17.9231
Epoch 184/200
4/4 [==============================] - 0s 1ms/step - loss: 18.0652
Epoch 185/200
4/4 [==============================] - 0s 2ms/step - loss: 17.9978
Epoch 186/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9539
Epoch 187/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9550
Epoch 188/200
4/4 [==============================] - 0s 1ms/step - loss: 17.8413
Epoch 189/200
4/4 [==============================] - 0s 1ms/step - loss: 18.0824
Epoch 190/200
4/4 [==============================] - 0s 1ms/step - loss: 17.8459
Epoch 191/200
4/4 [==============================] - 0s 1ms/step - loss: 17.7732
Epoch 192/200
4/4 [==============================] - 0s 1ms/step - loss: 17.7758
Epoch 193/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9182
Epoch 194/200
4/4 [==============================] - 0s 1ms/step - loss: 17.7538
Epoch 195/200
4/4 [==============================] - 0s 1ms/step - loss: 17.8182
Epoch 196/200
4/4 [==============================] - 0s 1ms/step - loss: 17.7276
Epoch 197/200
4/4 [==============================] - 0s 1ms/step - loss: 17.9054
Epoch 198/200
4/4 [==============================] - 0s 1ms/step - loss: 18.1024
Epoch 199/200
4/4 [==============================] - 0s 1ms/step - loss: 17.7250
Epoch 200/200
4/4 [==============================] - 0s 2ms/step - loss: 17.9618
<tensorflow.python.keras.callbacks.History at 0x1a4db1a3390>
loss = model.history.history['loss']
epochs = range(len(loss))
plt.plot(epochs, loss)
[<matplotlib.lines.Line2D at 0x1a4dcfc7320>]
x_for_predictions = np.linspace(0, 50, 100)
y_pred = model.predict(x_for_predictions)
y_pred
array([[ 2.7469127],
[ 3.77153 ],
[ 4.796148 ],
[ 5.820765 ],
[ 6.8453813],
[ 7.8699994],
[ 8.894616 ],
[ 9.919232 ],
[ 10.94385 ],
[ 11.968469 ],
[ 12.993085 ],
[ 14.017701 ],
[ 15.042319 ],
[ 16.066938 ],
[ 17.091555 ],
[ 18.116175 ],
[ 19.140789 ],
[ 20.165407 ],
[ 21.190025 ],
[ 22.214643 ],
[ 23.23926 ],
[ 24.263876 ],
[ 25.288492 ],
[ 26.313112 ],
[ 27.337729 ],
[ 28.362345 ],
[ 29.386963 ],
[ 30.411581 ],
[ 31.436197 ],
[ 32.46081 ],
[ 33.48543 ],
[ 34.510048 ],
[ 35.534664 ],
[ 36.559277 ],
[ 37.5839 ],
[ 38.608517 ],
[ 39.633137 ],
[ 40.65775 ],
[ 41.682373 ],
[ 42.70699 ],
[ 43.7316 ],
[ 44.75622 ],
[ 45.780834 ],
[ 46.805458 ],
[ 47.83007 ],
[ 48.854683 ],
[ 49.879303 ],
[ 50.90392 ],
[ 51.92854 ],
[ 52.953156 ],
[ 53.977776 ],
[ 55.00239 ],
[ 56.02701 ],
[ 57.05163 ],
[ 58.076244 ],
[ 59.10086 ],
[ 60.125477 ],
[ 61.15009 ],
[ 62.17471 ],
[ 63.199333 ],
[ 64.22395 ],
[ 65.248566 ],
[ 66.273186 ],
[ 67.2978 ],
[ 68.32242 ],
[ 69.34703 ],
[ 70.37166 ],
[ 71.39627 ],
[ 72.42089 ],
[ 73.4455 ],
[ 74.470116 ],
[ 75.49474 ],
[ 76.51936 ],
[ 77.543976 ],
[ 78.568596 ],
[ 79.59321 ],
[ 80.617836 ],
[ 81.64245 ],
[ 82.66706 ],
[ 83.69168 ],
[ 84.7163 ],
[ 85.74092 ],
[ 86.765526 ],
[ 87.79015 ],
[ 88.814766 ],
[ 89.83939 ],
[ 90.864006 ],
[ 91.88862 ],
[ 92.91323 ],
[ 93.93785 ],
[ 94.96246 ],
[ 95.98709 ],
[ 97.0117 ],
[ 98.03632 ],
[ 99.060936 ],
[100.085556 ],
[101.11017 ],
[102.134796 ],
[103.1594 ],
[104.18403 ]], dtype=float32)
plt.plot(x,y, '*')
plt.plot(x_for_predictions, y_pred, 'r')
[<matplotlib.lines.Line2D at 0x1a4dd057f28>]
from sklearn.metrics import mean_squared_error
from statsmodels.tools.eval_measures import mse
print(mean_squared_error(y,y_pred))
17.764145266496467
a = [1,2,3,4,5]
b = [2,3,4,5,6]
mean_squared_error(a,b)
1.0
mse(a,b)
1.0
y_pred_reshape = y_pred.reshape(100, )
y_pred_reshape.shape
(100,)
mse(y, y_pred_reshape)
17.764145266496467
Recurrent Neural Network
import pandas as pd
import numpy as np
%matplotlib inline
import matplotlib.pyplot as plt
df = pd.read_csv('Data/Alcohol_Sales.csv',index_col='DATE',parse_dates=True)
df.index.freq = 'MS'
df.head()
| S4248SM144NCEN | |
|---|---|
| DATE | |
| 1992-01-01 | 3459 |
| 1992-02-01 | 3458 |
| 1992-03-01 | 4002 |
| 1992-04-01 | 4564 |
| 1992-05-01 | 4221 |
df.columns = ['Sales']
df.plot(figsize=(12,8))
<AxesSubplot:xlabel='DATE'>
from statsmodels.tsa.seasonal import seasonal_decompose
results = seasonal_decompose(df['Sales'])
results.seasonal #plot(figsize=(12,8))
DATE
1992-01-01 -1836.088354
1992-02-01 -1321.328738
1992-03-01 -148.299892
1992-04-01 -240.372008
1992-05-01 554.966133
...
2018-09-01 -109.224572
2018-10-01 260.730557
2018-11-01 258.666454
2018-12-01 1303.892416
2019-01-01 -1836.088354
Freq: MS, Name: seasonal, Length: 325, dtype: float64
len(df)
325
nobs = 12
train = df.iloc[:-nobs]
test = df.iloc[-nobs:]
ser = np.array([23,56,2,13,14])
ser.max()
56
ser / ser.max()
array([0.41071429, 1. , 0.03571429, 0.23214286, 0.25 ])
from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()
scaler.fit(train) # finds the max value in train data
MinMaxScaler()
scaled_train = scaler.transform(train)
scaled_test = scaler.transform(test)
from tensorflow.keras.preprocessing.sequence import TimeseriesGenerator
n_input = 3
n_features = 1
generator = TimeseriesGenerator(scaled_train, scaled_train,length=n_input,batch_size=1)
len(scaled_train)
313
len(generator)
310
X,y = generator[0]
print(X)
print(y)
[[[0.03658432]
[0.03649885]
[0.08299855]]]
[[0.13103684]]
X.shape
(1, 3, 1)
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, LSTM
n_input = 12
n_features = 1
train_generator = TimeseriesGenerator(scaled_train, scaled_train,length=n_input,batch_size=1)
model = Sequential()
model.add(LSTM(150, activation = 'relu', input_shape=(n_input, n_features)))
model.add(Dense(1))
model.compile(optimizer='adam', loss='mse')
WARNING:tensorflow:Layer lstm_1 will not use cuDNN kernel since it doesn't meet the cuDNN kernel criteria. It will use generic GPU kernel as fallback when running on GPU
model.summary()
Model: "sequential_3"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
lstm_1 (LSTM) (None, 150) 91200
_________________________________________________________________
dense_7 (Dense) (None, 1) 151
=================================================================
Total params: 91,351
Trainable params: 91,351
Non-trainable params: 0
_________________________________________________________________
model.fit_generator(train_generator, epochs=25)
WARNING:tensorflow:From <ipython-input-83-698d8e58e24d>:1: Model.fit_generator (from tensorflow.python.keras.engine.training) is deprecated and will be removed in a future version.
Instructions for updating:
Please use Model.fit, which supports generators.
Epoch 1/25
301/301 [==============================] - 5s 18ms/step - loss: 0.0170
Epoch 2/25
301/301 [==============================] - 5s 18ms/step - loss: 0.0098
Epoch 3/25
301/301 [==============================] - 5s 18ms/step - loss: 0.0087
Epoch 4/25
301/301 [==============================] - 5s 18ms/step - loss: 0.0086
Epoch 5/25
301/301 [==============================] - 5s 18ms/step - loss: 0.0077
Epoch 6/25
301/301 [==============================] - 5s 15ms/step - loss: 0.0051
Epoch 7/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0042
Epoch 8/25
301/301 [==============================] - 5s 15ms/step - loss: 0.0036
Epoch 9/25
301/301 [==============================] - 5s 15ms/step - loss: 0.0028
Epoch 10/25
301/301 [==============================] - 5s 15ms/step - loss: 0.0022
Epoch 11/25
301/301 [==============================] - 5s 15ms/step - loss: 0.0024
Epoch 12/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0021
Epoch 13/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0021
Epoch 14/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0028: 0s - lo
Epoch 15/25
301/301 [==============================] - 5s 15ms/step - loss: 0.0018
Epoch 16/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0017: 0s - l
Epoch 17/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0017
Epoch 18/25
301/301 [==============================] - 5s 15ms/step - loss: 0.0017
Epoch 19/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0017
Epoch 20/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0018
Epoch 21/25
301/301 [==============================] - 5s 15ms/step - loss: 0.0017
Epoch 22/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0018
Epoch 23/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0016
Epoch 24/25
301/301 [==============================] - 5s 15ms/step - loss: 0.0018
Epoch 25/25
301/301 [==============================] - 5s 16ms/step - loss: 0.0015
<tensorflow.python.keras.callbacks.History at 0x1a5fca6e0f0>
model.history.history.keys()
dict_keys(['loss'])
plt.plot(range(len(model.history.history['loss'])),model.history.history['loss'])
[<matplotlib.lines.Line2D at 0x1a5ffd1dda0>]
# 12 history steps ---> step 13
# Last 12 points train --> point 1 of test data
first_eval_batch = scaled_train[-12:]
first_eval_batch
array([[0.63432772],
[0.80776135],
[0.72313873],
[0.89870929],
[1. ],
[0.71672793],
[0.88648602],
[0.75869732],
[0.82742115],
[0.87443371],
[0.96025301],
[0.5584238 ]])
first_eval_batch = first_eval_batch.reshape((1,n_input, n_features))
model.predict(first_eval_batch)
array([[0.7314653]], dtype=float32)
FORECAST USING RNN MODEL
test_predictions = []
first_eval_batch = scaled_train[-n_input:]
current_batch = first_eval_batch.reshape((1, n_input, n_features))
for i in range(len(test)):
current_pred = model.predict(current_batch)[0]
test_predictions.append(current_pred)
current_batch = np.append(current_batch[:,1:,:],[[current_pred]],axis=1)
test_predictions
[array([0.7314653], dtype=float32),
array([0.8324642], dtype=float32),
array([0.8028721], dtype=float32),
array([0.94515663], dtype=float32),
array([1.008543], dtype=float32),
array([0.79361135], dtype=float32),
array([0.9141835], dtype=float32),
array([0.8050848], dtype=float32),
array([0.87166405], dtype=float32),
array([0.91545624], dtype=float32),
array([0.97476], dtype=float32),
array([0.66045696], dtype=float32)]
true_predictions = scaler.inverse_transform(test_predictions)
true_predictions
array([[11588.41231138],
[12769.99888802],
[12423.80094749],
[14088.38745946],
[14829.94472694],
[12315.45915645],
[13726.03273642],
[12449.68736255],
[13228.59768867],
[13740.92249805],
[14434.71719247],
[10757.68592161]])
test['Predictions'] = true_predictions
C:\Users\ilvna\.conda\envs\tf-2.3\lib\site-packages\ipykernel_launcher.py:1: SettingWithCopyWarning:
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead
See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
"""Entry point for launching an IPython kernel.
test
| Sales | Predictions | |
|---|---|---|
| DATE | ||
| 2018-02-01 | 10415 | 11588.412311 |
| 2018-03-01 | 12683 | 12769.998888 |
| 2018-04-01 | 11919 | 12423.800947 |
| 2018-05-01 | 14138 | 14088.387459 |
| 2018-06-01 | 14583 | 14829.944727 |
| 2018-07-01 | 12640 | 12315.459156 |
| 2018-08-01 | 14257 | 13726.032736 |
| 2018-09-01 | 12396 | 12449.687363 |
| 2018-10-01 | 13914 | 13228.597689 |
| 2018-11-01 | 14174 | 13740.922498 |
| 2018-12-01 | 15504 | 14434.717192 |
| 2019-01-01 | 10718 | 10757.685922 |
test.plot(figsize=(12,8))
<AxesSubplot:xlabel='DATE'>
model.save('mycoolmodel.h5')
pwd
'C:\\Users\\ilvna\\Udemy'
from tensorflow.keras.models import load_model
new_model = load_model('mycoolmodel.h5')
WARNING:tensorflow:Layer lstm_1 will not use cuDNN kernel since it doesn't meet the cuDNN kernel criteria. It will use generic GPU kernel as fallback when running on GPU
new_model.summary()
Model: "sequential_3"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
lstm_1 (LSTM) (None, 150) 91200
_________________________________________________________________
dense_7 (Dense) (None, 1) 151
=================================================================
Total params: 91,351
Trainable params: 91,351
Non-trainable params: 0
_________________________________________________________________