Stock Prediction using Time Series Analysis

Closing Price prediction of Yahoo stocks from 2010 - 2016 using Gated Recurrant Units	
Model is already trained and saved in 'stock_price_GRU.h5' file	
To obtain the trained model just comment out the lines 47-55 and 60-62, then uncomment the lines 57-58 to load 'stock_price_GRU.h5' file	

Highly Recommend using GPU version of Tensorflow for running the model	

DATA

INPUT_DATA
date             open        low       high      close
2010-01-04  16.940001  16.879999  17.200001  17.100000
2010-01-05  17.219999  17.000000  17.230000  17.230000
2010-01-06  17.170000  17.070000  17.299999  17.170000
2010-01-07  16.809999  16.570000  16.900000  16.700001
2010-01-08  16.680000  16.620001  16.760000  16.700001

LABEL_DATA
date		  close
2010-01-04    17.230000
2010-01-05    17.170000
2010-01-06    16.700001
2010-01-07    16.700001
2010-01-08    16.740000

MODEL

Layer (type) Output Shape Param #

---

gru_1 (GRU) (None, 1, 512) 794112

---

dropout_1 (Dropout) (None, 1, 512) 0

---

gru_2 (GRU) (None, 256) 590592

---

dropout_2 (Dropout) (None, 256) 0

---

dense_1 (Dense) (None, 1) 257

---

Total params: 1,384,961 Trainable params: 1,384,961 Non-trainable params: 0

---

TRAINING

Epoch 500/500
250/1061 [======>.......................] - ETA: 0s - loss: 7.2934e-04
750/1061 [====================>.........] - ETA: 0s - loss: 6.7267e-04
1061/1061 [==============================] - 0s 111us/step - loss: 6.4617e-04 - val_loss: 6.4601e-04

32/582 [>.............................] - ETA: 0s
352/582 [=================>............] - ETA: 0s
582/582 [==============================] - 0s 154us/step
Score: 0.000513115886573222	

RESULTS

33% of Data used for Testing 
Plot only shows the last points of test set and predicted values