Recurrent Neural Networks Part 3

June 26, 2019

This week we will be covering the final part of RNN development to predict stock price information using previous values.

This notebook is based on a tutorial: “Multivariate Time Series using RNN with Keras” from here. We have modified the approach to use all available values as input instead of averaging the low and high price.

View the IPython notebook for this session on Github [here](https://github.com/HSV-AI/presentations/blob/master//2019/190626_RNN_TimeSeries_Part3.ipynb

Or launch the notebook in Google Colab or MyBinder:

Meeting Agenda

Introduction
Project Updates
RNN Code Walkthrough
AI Model Deployment

RNN Example Code

import numpy as np
import pandas as pd
import math
from keras import Sequential
from keras.layers import Dense, LSTM
from sklearn.preprocessing import MinMaxScaler
import matplotlib.pyplot as plt
%matplotlib inline

from sklearn.model_selection import train_test_split
total_training = []
total_validation = []

First, we download a csv export of the Google stock from here.

For convenience, this is saved in the presentations repository and can be access by URL.

Note:

We need to use the thousands=‘,’ parameter in order to ignore the commas in the numbers when converting to floats.

We’ll use the head() method to get a look at what the dataframe looks like.

fileURL = "https://github.com/HSV-AI/presentations/raw/master/2019/190619/HistoricalQuotes.csv"
stock_data = pd.read_csv(fileURL, thousands=',')
stock_data.head(5)

	date	close	volume	open	high	low
0	16:00	1103.60	1386584.0	1107.82	1116.39	1098.9900
1	2019/06/18	1103.60	1386684.0	1109.69	1116.39	1098.9900
2	2019/06/17	1092.50	941602.0	1086.28	1099.18	1086.2800
3	2019/06/14	1085.35	1111643.0	1086.42	1092.69	1080.1721
4	2019/06/13	1088.77	1058000.0	1083.64	1094.17	1080.1500

stock_data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1260 entries, 0 to 1259
Data columns (total 6 columns):
date      1260 non-null object
close     1260 non-null float64
volume    1260 non-null float64
open      1260 non-null float64
high      1260 non-null float64
low       1260 non-null float64
dtypes: float64(5), object(1)
memory usage: 59.1+ KB

The example showed how to use 2 values as input – the volume and an average of the high/low price. I’ve changed this to just use the 5 values individually and let the network figure out how they might be related when predicting an output.

input_feature= stock_data.iloc[:,1:].values
input_data = input_feature
input_feature.shape

output_feature = stock_data.iloc[:,[1,3]].values

print(input_feature.shape)
print(output_feature.shape)
stock_data.iloc[:,1:].head(5)

(1260, 5)
(1260, 2)

	close	volume	open	high	low
0	1103.60	1386584.0	1107.82	1116.39	1098.9900
1	1103.60	1386684.0	1109.69	1116.39	1098.9900
2	1092.50	941602.0	1086.28	1099.18	1086.2800
3	1085.35	1111643.0	1086.42	1092.69	1080.1721
4	1088.77	1058000.0	1083.64	1094.17	1080.1500

plt.figure(figsize=(10,8))
plt.plot(input_feature[:,1])
plt.title("Volume of stocks sold")
plt.xlabel("Time (latest-> oldest)")
plt.ylabel("Volume of stocks traded")
plt.show()

plt.figure(figsize=(10,8))
plt.plot(input_feature[:,2], color='blue')
plt.title("Google Stock Prices")
plt.xlabel("Time (latest-> oldest)")
plt.ylabel("Stock Open Price")
plt.show()

input_scaler = MinMaxScaler(feature_range=(0,1))
input_data[:,:] = input_scaler.fit_transform(input_feature[:,:])

output_scaler = MinMaxScaler(feature_range=(0,1))
# output_feature = output_feature.reshape(-1, 1)
output_data = output_scaler.fit_transform(output_feature[:,:])

lookback= 50

X=[]
y=[]
for i in range(len(stock_data)-lookback-1):
    t=[]
    for j in range(0,lookback):
        t.append(input_data[[(i+j)], :])

    X.append(t)
    y.append(output_data[i+ lookback])

X, y= np.array(X), np.array(y)
X = X.reshape(X.shape[0],lookback, 5)

print(X.shape)
print(y.shape)

(1209, 50, 5)
(1209, 2)

model = Sequential()
model.add(LSTM(units=30, return_sequences= True, input_shape=(X.shape[1],5)))
model.add(LSTM(units=30, return_sequences=True))
model.add(LSTM(units=30))
model.add(Dense(units=2))
model.summary()

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
lstm_4 (LSTM)                (None, 50, 30)            4320
_________________________________________________________________
lstm_5 (LSTM)                (None, 50, 30)            7320
_________________________________________________________________
lstm_6 (LSTM)                (None, 30)                7320
_________________________________________________________________
dense_2 (Dense)              (None, 2)                 62
=================================================================
Total params: 19,022
Trainable params: 19,022
Non-trainable params: 0
_________________________________________________________________

model.compile(optimizer='adam', loss='mean_squared_error')

W0626 22:12:08.926577 140698563360640 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/optimizers.py:790: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.

from IPython.display import SVG
from keras.utils.vis_utils import model_to_dot

SVG(model_to_dot(model, show_shapes=True).create(prog='dot', format='svg'))

def plot_loss(history):

  length = len(history.history['loss'])
  rng = range(length)

  training_fit = np.polyfit(rng, history.history['loss'], 1)
  training_trend = np.poly1d(training_fit)

  val_fit = np.polyfit(rng, history.history['val_loss'], 1)
  val_trend = np.poly1d(val_fit)

  plt.figure(figsize=(10,8))
  plt.plot(history.history['loss'], color='green')
  plt.plot(training_trend(rng), color='green', linestyle='--')
  plt.plot(history.history['val_loss'], color='orange')
  plt.plot(val_trend(rng), color='orange', linestyle='--')

  plt.title('Model loss')
  plt.ylabel('Loss')
  plt.xlabel('Epoch')
  plt.legend(['Training', 'Training Trend', 'Validation', 'Validation Trend'], loc='lower left')
  plt.show()

Originally, we had this line:

history = model.fit(X, y, validation_split=0.30, epochs=50, batch_size=32, verbose=1)

The problem is that the validation split always takes the validation off of the back of the data, instead of randomly selecting 30% for validation.

To change this up, we will use the train_test_split from sklearn to give us a random 30% validation set. Then we will train for 50 epochs and graph the training loss and validation loss to see how we’re doing.

We will repeat this for a few times to see how low the error drops.

for i in range(4):
  XTraining, XValidation, YTraining, YValidation = train_test_split(X,y,test_size=0.3) # before model building
  history = model.fit(XTraining, YTraining, batch_size=32, epochs=50, validation_data=(XValidation,YValidation))
  total_training.extend(history.history['loss'])
  total_validation.extend(history.history['val_loss'])

  # Plot training & validation loss values
  plot_loss(history)

W0626 22:12:19.481230 140698563360640 deprecation.py:323] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/math_grad.py:1250: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.where in 2.0, which has the same broadcast rule as np.where
W0626 22:12:21.329057 140698563360640 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:986: The name tf.assign_add is deprecated. Please use tf.compat.v1.assign_add instead.

W0626 22:12:21.484905 140698563360640 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:973: The name tf.assign is deprecated. Please use tf.compat.v1.assign instead.



Train on 846 samples, validate on 363 samples
Epoch 1/50
846/846 [==============================] - 5s 6ms/step - loss: 0.0493 - val_loss: 0.0074
Epoch 2/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0036 - val_loss: 0.0018
Epoch 3/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0019 - val_loss: 0.0015
Epoch 4/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0017 - val_loss: 0.0015
Epoch 5/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0017 - val_loss: 0.0014
Epoch 6/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0016 - val_loss: 0.0013
Epoch 7/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0016 - val_loss: 0.0017
Epoch 8/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0017 - val_loss: 0.0018
Epoch 9/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0015 - val_loss: 0.0013
Epoch 10/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0015 - val_loss: 0.0013
Epoch 11/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0015 - val_loss: 0.0012
Epoch 12/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0014 - val_loss: 0.0013
Epoch 13/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0014 - val_loss: 0.0011
Epoch 14/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0014 - val_loss: 0.0011
Epoch 15/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0014 - val_loss: 0.0012
Epoch 16/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0013 - val_loss: 0.0012
Epoch 17/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0013 - val_loss: 0.0010
Epoch 18/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0012 - val_loss: 0.0010
Epoch 19/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0014 - val_loss: 9.9284e-04
Epoch 20/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0013 - val_loss: 9.9700e-04
Epoch 21/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0013 - val_loss: 0.0012
Epoch 22/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0014 - val_loss: 0.0012
Epoch 23/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0012 - val_loss: 9.9032e-04
Epoch 24/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0012 - val_loss: 9.3364e-04
Epoch 25/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0011 - val_loss: 9.2768e-04
Epoch 26/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0012 - val_loss: 9.4095e-04
Epoch 27/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0011 - val_loss: 9.4460e-04
Epoch 28/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0011 - val_loss: 0.0010
Epoch 29/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0012 - val_loss: 9.3908e-04
Epoch 30/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0013 - val_loss: 9.7271e-04
Epoch 31/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0012 - val_loss: 9.4462e-04
Epoch 32/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0012 - val_loss: 8.6673e-04
Epoch 33/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0011 - val_loss: 8.4075e-04
Epoch 34/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0010 - val_loss: 8.4946e-04
Epoch 35/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0010 - val_loss: 8.4546e-04
Epoch 36/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0010 - val_loss: 9.0856e-04
Epoch 37/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0010 - val_loss: 8.2692e-04
Epoch 38/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0010 - val_loss: 8.6757e-04
Epoch 39/50
846/846 [==============================] - 2s 3ms/step - loss: 9.7241e-04 - val_loss: 8.8182e-04
Epoch 40/50
846/846 [==============================] - 2s 3ms/step - loss: 9.8657e-04 - val_loss: 8.1574e-04
Epoch 41/50
846/846 [==============================] - 2s 3ms/step - loss: 9.7507e-04 - val_loss: 8.6977e-04
Epoch 42/50
846/846 [==============================] - 2s 3ms/step - loss: 9.3582e-04 - val_loss: 7.9621e-04
Epoch 43/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0010 - val_loss: 8.7585e-04
Epoch 44/50
846/846 [==============================] - 2s 3ms/step - loss: 9.8852e-04 - val_loss: 8.0546e-04
Epoch 45/50
846/846 [==============================] - 2s 3ms/step - loss: 0.0011 - val_loss: 7.3085e-04
Epoch 46/50
846/846 [==============================] - 2s 3ms/step - loss: 9.0648e-04 - val_loss: 7.8457e-04
Epoch 47/50
846/846 [==============================] - 2s 3ms/step - loss: 9.5309e-04 - val_loss: 7.5034e-04
Epoch 48/50
846/846 [==============================] - 2s 3ms/step - loss: 8.5657e-04 - val_loss: 8.2050e-04
Epoch 49/50
846/846 [==============================] - 2s 3ms/step - loss: 8.1200e-04 - val_loss: 9.5116e-04
Epoch 50/50
846/846 [==============================] - 2s 3ms/step - loss: 8.7380e-04 - val_loss: 7.0381e-04

Train on 846 samples, validate on 363 samples
Epoch 1/50
846/846 [==============================] - 2s 3ms/step - loss: 7.8567e-04 - val_loss: 8.3607e-04
Epoch 2/50
846/846 [==============================] - 2s 3ms/step - loss: 8.8357e-04 - val_loss: 8.1796e-04
Epoch 3/50
846/846 [==============================] - 2s 3ms/step - loss: 8.3822e-04 - val_loss: 7.4931e-04
Epoch 4/50
846/846 [==============================] - 2s 3ms/step - loss: 8.2381e-04 - val_loss: 8.2596e-04
Epoch 5/50
846/846 [==============================] - 2s 3ms/step - loss: 7.6646e-04 - val_loss: 7.2589e-04
Epoch 6/50
846/846 [==============================] - 2s 3ms/step - loss: 8.4898e-04 - val_loss: 8.1355e-04
Epoch 7/50
846/846 [==============================] - 2s 3ms/step - loss: 8.9905e-04 - val_loss: 7.0409e-04
Epoch 8/50
846/846 [==============================] - 2s 3ms/step - loss: 8.4297e-04 - val_loss: 7.0732e-04
Epoch 9/50
846/846 [==============================] - 2s 3ms/step - loss: 7.1665e-04 - val_loss: 6.9534e-04
Epoch 10/50
846/846 [==============================] - 2s 3ms/step - loss: 7.2126e-04 - val_loss: 6.8278e-04
Epoch 11/50
846/846 [==============================] - 2s 3ms/step - loss: 7.1083e-04 - val_loss: 6.7135e-04
Epoch 12/50
846/846 [==============================] - 2s 3ms/step - loss: 7.6266e-04 - val_loss: 7.1941e-04
Epoch 13/50
846/846 [==============================] - 2s 3ms/step - loss: 7.5223e-04 - val_loss: 7.6919e-04
Epoch 14/50
846/846 [==============================] - 2s 3ms/step - loss: 7.4737e-04 - val_loss: 6.9894e-04
Epoch 15/50
846/846 [==============================] - 2s 3ms/step - loss: 7.6271e-04 - val_loss: 6.2322e-04
Epoch 16/50
846/846 [==============================] - 2s 3ms/step - loss: 7.0625e-04 - val_loss: 6.1616e-04
Epoch 17/50
846/846 [==============================] - 2s 3ms/step - loss: 8.3463e-04 - val_loss: 6.5984e-04
Epoch 18/50
846/846 [==============================] - 2s 3ms/step - loss: 7.2503e-04 - val_loss: 5.9163e-04
Epoch 19/50
846/846 [==============================] - 2s 3ms/step - loss: 6.4949e-04 - val_loss: 5.8263e-04
Epoch 20/50
846/846 [==============================] - 2s 3ms/step - loss: 7.3382e-04 - val_loss: 5.5372e-04
Epoch 21/50
846/846 [==============================] - 2s 3ms/step - loss: 6.4378e-04 - val_loss: 5.9549e-04
Epoch 22/50
846/846 [==============================] - 2s 3ms/step - loss: 6.1992e-04 - val_loss: 6.1535e-04
Epoch 23/50
846/846 [==============================] - 2s 3ms/step - loss: 6.9907e-04 - val_loss: 5.8023e-04
Epoch 24/50
846/846 [==============================] - 2s 3ms/step - loss: 6.2607e-04 - val_loss: 6.4950e-04
Epoch 25/50
846/846 [==============================] - 2s 3ms/step - loss: 6.8107e-04 - val_loss: 5.5166e-04
Epoch 26/50
846/846 [==============================] - 2s 3ms/step - loss: 6.0072e-04 - val_loss: 5.2261e-04
Epoch 27/50
846/846 [==============================] - 2s 3ms/step - loss: 5.6027e-04 - val_loss: 4.7815e-04
Epoch 28/50
846/846 [==============================] - 2s 3ms/step - loss: 5.5287e-04 - val_loss: 4.8670e-04
Epoch 29/50
846/846 [==============================] - 2s 3ms/step - loss: 5.4715e-04 - val_loss: 4.5585e-04
Epoch 30/50
846/846 [==============================] - 2s 3ms/step - loss: 5.5927e-04 - val_loss: 4.5167e-04
Epoch 31/50
846/846 [==============================] - 2s 3ms/step - loss: 5.5716e-04 - val_loss: 4.9055e-04
Epoch 32/50
846/846 [==============================] - 2s 3ms/step - loss: 5.4521e-04 - val_loss: 5.1821e-04
Epoch 33/50
846/846 [==============================] - 2s 3ms/step - loss: 5.5661e-04 - val_loss: 5.1382e-04
Epoch 34/50
846/846 [==============================] - 2s 3ms/step - loss: 5.8093e-04 - val_loss: 6.5531e-04
Epoch 35/50
846/846 [==============================] - 2s 3ms/step - loss: 6.3154e-04 - val_loss: 4.4359e-04
Epoch 36/50
846/846 [==============================] - 2s 3ms/step - loss: 5.5163e-04 - val_loss: 7.0861e-04
Epoch 37/50
846/846 [==============================] - 2s 3ms/step - loss: 6.6860e-04 - val_loss: 6.9380e-04
Epoch 38/50
846/846 [==============================] - 2s 3ms/step - loss: 5.9600e-04 - val_loss: 4.0907e-04
Epoch 39/50
846/846 [==============================] - 2s 3ms/step - loss: 5.0720e-04 - val_loss: 3.8786e-04
Epoch 40/50
846/846 [==============================] - 2s 3ms/step - loss: 4.5425e-04 - val_loss: 3.9124e-04
Epoch 41/50
846/846 [==============================] - 2s 3ms/step - loss: 4.5075e-04 - val_loss: 3.6695e-04
Epoch 42/50
846/846 [==============================] - 2s 3ms/step - loss: 4.9808e-04 - val_loss: 5.3683e-04
Epoch 43/50
846/846 [==============================] - 2s 3ms/step - loss: 5.0339e-04 - val_loss: 4.0243e-04
Epoch 44/50
846/846 [==============================] - 2s 3ms/step - loss: 4.3682e-04 - val_loss: 3.5072e-04
Epoch 45/50
846/846 [==============================] - 2s 3ms/step - loss: 4.2299e-04 - val_loss: 3.4365e-04
Epoch 46/50
846/846 [==============================] - 2s 3ms/step - loss: 4.1534e-04 - val_loss: 3.3323e-04
Epoch 47/50
846/846 [==============================] - 2s 3ms/step - loss: 4.4758e-04 - val_loss: 3.5490e-04
Epoch 48/50
846/846 [==============================] - 2s 3ms/step - loss: 4.2959e-04 - val_loss: 3.3163e-04
Epoch 49/50
846/846 [==============================] - 2s 3ms/step - loss: 3.8745e-04 - val_loss: 3.2710e-04
Epoch 50/50
846/846 [==============================] - 2s 3ms/step - loss: 4.1643e-04 - val_loss: 3.9325e-04

Train on 846 samples, validate on 363 samples
Epoch 1/50
846/846 [==============================] - 2s 3ms/step - loss: 3.6714e-04 - val_loss: 4.1345e-04
Epoch 2/50
846/846 [==============================] - 2s 3ms/step - loss: 3.5691e-04 - val_loss: 3.5829e-04
Epoch 3/50
846/846 [==============================] - 2s 3ms/step - loss: 3.4347e-04 - val_loss: 3.9090e-04
Epoch 4/50
846/846 [==============================] - 2s 3ms/step - loss: 3.6989e-04 - val_loss: 3.7556e-04
Epoch 5/50
846/846 [==============================] - 2s 3ms/step - loss: 3.6009e-04 - val_loss: 3.4568e-04
Epoch 6/50
846/846 [==============================] - 2s 3ms/step - loss: 3.6599e-04 - val_loss: 3.3915e-04
Epoch 7/50
846/846 [==============================] - 2s 3ms/step - loss: 3.4199e-04 - val_loss: 3.4132e-04
Epoch 8/50
846/846 [==============================] - 2s 3ms/step - loss: 3.1425e-04 - val_loss: 3.2638e-04
Epoch 9/50
846/846 [==============================] - 2s 3ms/step - loss: 3.4761e-04 - val_loss: 3.7415e-04
Epoch 10/50
846/846 [==============================] - 2s 3ms/step - loss: 3.2125e-04 - val_loss: 3.1974e-04
Epoch 11/50
846/846 [==============================] - 2s 3ms/step - loss: 2.9812e-04 - val_loss: 3.1873e-04
Epoch 12/50
846/846 [==============================] - 2s 3ms/step - loss: 2.9344e-04 - val_loss: 3.3415e-04
Epoch 13/50
846/846 [==============================] - 2s 3ms/step - loss: 2.9954e-04 - val_loss: 3.7027e-04
Epoch 14/50
846/846 [==============================] - 2s 3ms/step - loss: 2.8476e-04 - val_loss: 4.0148e-04
Epoch 15/50
846/846 [==============================] - 2s 3ms/step - loss: 3.0628e-04 - val_loss: 3.2988e-04
Epoch 16/50
846/846 [==============================] - 2s 3ms/step - loss: 2.8418e-04 - val_loss: 4.0104e-04
Epoch 17/50
846/846 [==============================] - 2s 3ms/step - loss: 2.9285e-04 - val_loss: 3.4057e-04
Epoch 18/50
846/846 [==============================] - 2s 3ms/step - loss: 2.8126e-04 - val_loss: 2.9773e-04
Epoch 19/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5628e-04 - val_loss: 3.9303e-04
Epoch 20/50
846/846 [==============================] - 2s 3ms/step - loss: 3.2218e-04 - val_loss: 3.4024e-04
Epoch 21/50
846/846 [==============================] - 2s 3ms/step - loss: 2.7909e-04 - val_loss: 2.8494e-04
Epoch 22/50
846/846 [==============================] - 2s 3ms/step - loss: 2.7626e-04 - val_loss: 2.8021e-04
Epoch 23/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4787e-04 - val_loss: 2.9391e-04
Epoch 24/50
846/846 [==============================] - 2s 3ms/step - loss: 2.7745e-04 - val_loss: 3.0617e-04
Epoch 25/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4422e-04 - val_loss: 2.8924e-04
Epoch 26/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4333e-04 - val_loss: 3.1980e-04
Epoch 27/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5234e-04 - val_loss: 2.7220e-04
Epoch 28/50
846/846 [==============================] - 2s 3ms/step - loss: 2.7631e-04 - val_loss: 2.7548e-04
Epoch 29/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2936e-04 - val_loss: 2.8906e-04
Epoch 30/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5533e-04 - val_loss: 3.1416e-04
Epoch 31/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3497e-04 - val_loss: 3.3216e-04
Epoch 32/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5406e-04 - val_loss: 2.6688e-04
Epoch 33/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3960e-04 - val_loss: 3.0699e-04
Epoch 34/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3831e-04 - val_loss: 2.7836e-04
Epoch 35/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3423e-04 - val_loss: 2.8158e-04
Epoch 36/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2899e-04 - val_loss: 2.6425e-04
Epoch 37/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4251e-04 - val_loss: 2.6722e-04
Epoch 38/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1946e-04 - val_loss: 3.4129e-04
Epoch 39/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5077e-04 - val_loss: 4.2376e-04
Epoch 40/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5585e-04 - val_loss: 4.2111e-04
Epoch 41/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4775e-04 - val_loss: 2.8553e-04
Epoch 42/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2740e-04 - val_loss: 2.6651e-04
Epoch 43/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1321e-04 - val_loss: 2.9056e-04
Epoch 44/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3993e-04 - val_loss: 2.7909e-04
Epoch 45/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1702e-04 - val_loss: 3.9789e-04
Epoch 46/50
846/846 [==============================] - 2s 3ms/step - loss: 3.0304e-04 - val_loss: 4.4667e-04
Epoch 47/50
846/846 [==============================] - 2s 3ms/step - loss: 2.8742e-04 - val_loss: 2.9968e-04
Epoch 48/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2063e-04 - val_loss: 4.2202e-04
Epoch 49/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5344e-04 - val_loss: 2.8836e-04
Epoch 50/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1879e-04 - val_loss: 2.7415e-04

Train on 846 samples, validate on 363 samples
Epoch 1/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3371e-04 - val_loss: 2.1271e-04
Epoch 2/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4240e-04 - val_loss: 2.1845e-04
Epoch 3/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4028e-04 - val_loss: 2.1864e-04
Epoch 4/50
846/846 [==============================] - 2s 3ms/step - loss: 2.6314e-04 - val_loss: 2.1681e-04
Epoch 5/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4452e-04 - val_loss: 2.4099e-04
Epoch 6/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3376e-04 - val_loss: 2.3247e-04
Epoch 7/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5632e-04 - val_loss: 2.1544e-04
Epoch 8/50
846/846 [==============================] - 2s 3ms/step - loss: 2.6193e-04 - val_loss: 2.0350e-04
Epoch 9/50
846/846 [==============================] - 2s 3ms/step - loss: 3.4183e-04 - val_loss: 2.7302e-04
Epoch 10/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5634e-04 - val_loss: 2.6461e-04
Epoch 11/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2919e-04 - val_loss: 2.0459e-04
Epoch 12/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3505e-04 - val_loss: 2.0448e-04
Epoch 13/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3275e-04 - val_loss: 2.3731e-04
Epoch 14/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2792e-04 - val_loss: 2.4049e-04
Epoch 15/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2646e-04 - val_loss: 2.2522e-04
Epoch 16/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5407e-04 - val_loss: 2.1497e-04
Epoch 17/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3197e-04 - val_loss: 2.3586e-04
Epoch 18/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3642e-04 - val_loss: 2.1039e-04
Epoch 19/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2678e-04 - val_loss: 2.3727e-04
Epoch 20/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4133e-04 - val_loss: 2.2049e-04
Epoch 21/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3942e-04 - val_loss: 2.0113e-04
Epoch 22/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4385e-04 - val_loss: 2.8255e-04
Epoch 23/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4870e-04 - val_loss: 2.5603e-04
Epoch 24/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2633e-04 - val_loss: 2.1703e-04
Epoch 25/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2453e-04 - val_loss: 3.2252e-04
Epoch 26/50
846/846 [==============================] - 2s 3ms/step - loss: 2.4094e-04 - val_loss: 2.3855e-04
Epoch 27/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3521e-04 - val_loss: 2.3769e-04
Epoch 28/50
846/846 [==============================] - 2s 3ms/step - loss: 3.3634e-04 - val_loss: 5.2989e-04
Epoch 29/50
846/846 [==============================] - 2s 3ms/step - loss: 2.8109e-04 - val_loss: 2.3363e-04
Epoch 30/50
846/846 [==============================] - 2s 3ms/step - loss: 2.5225e-04 - val_loss: 1.9715e-04
Epoch 31/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3162e-04 - val_loss: 2.2016e-04
Epoch 32/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2651e-04 - val_loss: 2.5741e-04
Epoch 33/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2497e-04 - val_loss: 2.1622e-04
Epoch 34/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3961e-04 - val_loss: 2.0696e-04
Epoch 35/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1565e-04 - val_loss: 1.9688e-04
Epoch 36/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1195e-04 - val_loss: 1.9870e-04
Epoch 37/50
846/846 [==============================] - 2s 3ms/step - loss: 2.0837e-04 - val_loss: 2.0992e-04
Epoch 38/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3505e-04 - val_loss: 2.0513e-04
Epoch 39/50
846/846 [==============================] - 3s 3ms/step - loss: 2.6068e-04 - val_loss: 1.9653e-04
Epoch 40/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1517e-04 - val_loss: 1.9995e-04
Epoch 41/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1684e-04 - val_loss: 2.2420e-04
Epoch 42/50
846/846 [==============================] - 2s 3ms/step - loss: 2.2658e-04 - val_loss: 1.9264e-04
Epoch 43/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1324e-04 - val_loss: 2.3070e-04
Epoch 44/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3143e-04 - val_loss: 1.9386e-04
Epoch 45/50
846/846 [==============================] - 2s 3ms/step - loss: 2.3191e-04 - val_loss: 1.9974e-04
Epoch 46/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1504e-04 - val_loss: 2.6093e-04
Epoch 47/50
846/846 [==============================] - 2s 3ms/step - loss: 2.8914e-04 - val_loss: 4.3036e-04
Epoch 48/50
846/846 [==============================] - 2s 3ms/step - loss: 2.8469e-04 - val_loss: 1.9441e-04
Epoch 49/50
846/846 [==============================] - 2s 3ms/step - loss: 2.1440e-04 - val_loss: 1.9236e-04
Epoch 50/50
846/846 [==============================] - 2s 3ms/step - loss: 2.0695e-04 - val_loss: 2.3324e-04

Now we’ll take a look at the final results of the training/validation error. We throw out the first 10 epochs of training, as they vary wildly before any real training begins.

length = len(total_training[10:-1])
rng = range(length)

training_fit = np.polyfit(rng, total_training[10:-1], 4)
training_trend = np.poly1d(training_fit)

val_fit = np.polyfit(rng, total_validation[10:-1], 4)
val_trend = np.poly1d(val_fit)

plt.figure(figsize=(10,8))
plt.plot(total_training[10:-1], color='green')
plt.plot(training_trend(rng), color='green', linestyle='--')
plt.plot(total_validation[10:-1], color='orange')
plt.plot(val_trend(rng), color='orange', linestyle='--')

plt.title('Model loss')
plt.ylabel('Loss')
plt.xlabel('Epoch')
plt.legend(['Training', 'Training Trend', 'Validation', 'Validation Trend'], loc='upper right')
plt.show()

predicted_value = model.predict(X)
unscaled_output = output_scaler.inverse_transform(predicted_value)
print(predicted_value.shape)

temp = np.array(stock_data.open[lookback:-1]).reshape(-1,1)

(1209, 2)


plt.figure(figsize=(10,8))
plt.plot(unscaled_output[:,0], color= 'red')
plt.plot(temp, color='green')
plt.title("Opening price of stocks sold")
plt.xlabel("Time (latest-> oldest)")
plt.ylabel("Stock Opening Price")
plt.show()