# Import libraries
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import *
from tensorflow.keras.callbacks import ModelCheckpoint
from tensorflow.keras.losses import MeanSquaredError
from tensorflow.keras.optimizers import Adam
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error

from tensorflow.keras.losses import MeanSquaredError
from tensorflow.keras.metrics import RootMeanSquaredError
from tensorflow.keras.models import load_model

data = pd.read_csv('data/yahoo_stock.csv')

# Make the Date time column as the index
data.index = pd.to_datetime(data['Date'])
data.head(3)

# information of data
data.info()

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 1825 entries, 2015-11-23 to 2020-11-20
Data columns (total 7 columns):
 #   Column     Non-Null Count  Dtype  
---  ------     --------------  -----  
 0   Date       1825 non-null   object 
 1   High       1825 non-null   float64
 2   Low        1825 non-null   float64
 3   Open       1825 non-null   float64
 4   Close      1825 non-null   float64
 5   Volume     1825 non-null   float64
 6   Adj Close  1825 non-null   float64
dtypes: float64(6), object(1)
memory usage: 114.1+ KB

data.describe()

# Data Correlation: Relationship between columns
data[['Open','Close','Adj Close','High','Low']].corr()

# Visualization of correlation result with seaborn library heatmap.
f, ax = plt.subplots(figsize = (6,6))
sns.heatmap(data[['Open','Close','Adj Close','High','Low']].corr(), annot = True, linewidths=0.5, linecolor = "black", fmt = ".3f", ax = ax)
plt.show()

date = data.index

open_price = data.loc[:, ["Open"]]

plt.plot(date, open_price, label="opening prices")
plt.xlabel("Date")
plt.ylabel("Stock Prices")
plt.legend()
plt.title('Opening Stock Prices')
plt.show()

data = data.loc[:, ["Open"]].values

# Feature Scaling
sc = MinMaxScaler(feature_range=(0,1))
data = sc.fit_transform(data)

train_size = int(len(data) * 0.70)
remainder = round((len(data) - train_size)/2)
val_size = train_size + remainder

print(f"train_size === {train_size}; val_size and test_size === {val_size}")

train = data[0:train_size, :]
val = data[train_size:val_size, :]
test = data[val_size:len(data), :]
#test = data[train_size:len(data), :]
print(f"length of train === {len(train)}; length of val == {len(val)}; length of test === {len(test)}")

train_size === 1277; val_size and test_size === 1551
length of train === 1277; length of val == 274; length of test === 274

def create_dataset(dataset, time_steps):
    dataX = []
    dataY = []
    for i in range(len(dataset) - time_steps - 1):
        a = dataset[i: (i + time_steps), 0]
        dataX.append(a)
        dataY.append(dataset[i + time_steps, 0])
    return np.array(dataX), np.array(dataY)

# reshape into X=t and Y=t+1
time_steps = 50
n_features = 1 
X_train, y_train = create_dataset(dataset=train, time_steps=time_steps)
X_val, y_val = create_dataset(dataset=val, time_steps=time_steps)
X_test, y_test = create_dataset(dataset=test, time_steps=time_steps)
print(f"X_train shape === {X_train.shape}; y_train shape === {y_train.shape}")
print(f"X_val shape === {X_val.shape}; y_val shape === {y_val.shape}")
print(f"X_test shape === {X_test.shape}; y_test shape === {y_test.shape}")

X_train shape === (1226, 50); y_train shape === (1226,)
X_val shape === (223, 50); y_val shape === (223,)
X_test shape === (223, 50); y_test shape === (223,)

# Initialize LSTM network
model = Sequential()
# Add 1st layer LSTM and some Dropout regularisation
model.add(LSTM(units=60, return_sequences=True, input_shape=(time_steps, n_features)))
model.add(Dropout(0.1))
# Add 2nd layer LSTM and some Dropout regularisation
#model.add(LSTM(units= 60, return_sequences=True))
#model.add(Dropout(0.1))
# Add a third LSTM layer and some Dropout regularisation
model.add(LSTM(units= 60, return_sequences=False))
model.add(Dropout(0.1))
# Add the output layer
model.add(Dense(1))

#model summary
model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 lstm (LSTM)                 (None, 50, 60)            14880     
                                                                 
 dropout (Dropout)           (None, 50, 60)            0         
                                                                 
 lstm_1 (LSTM)               (None, 60)                29040     
                                                                 
 dropout_1 (Dropout)         (None, 60)                0         
                                                                 
 dense (Dense)               (None, 1)                 61        
                                                                 
=================================================================
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 lstm (LSTM)                 (None, 50, 60)            14880     
                                                                 
 dropout (Dropout)           (None, 50, 60)            0         
                                                                 
 lstm_1 (LSTM)               (None, 60)                29040     
                                                                 
 dropout_1 (Dropout)         (None, 60)                0         
                                                                 
 dense (Dense)               (None, 1)                 61        
                                                                 
=================================================================
Total params: 43,981
Trainable params: 43,981
Non-trainable params: 0
_________________________________________________________________

cp = ModelCheckpoint('ysp_models/', save_best_only=True)
model.compile(loss=MeanSquaredError(), optimizer=Adam(learning_rate=0.001), metrics=[RootMeanSquaredError()])

model.fit(X_train, y_train, validation_data=(X_val, y_val), epochs=100, batch_size=32, verbose=2, callbacks=[cp])

Epoch 1/100

WARNING:absl:Found untraced functions such as _update_step_xla, lstm_cell_layer_call_fn, lstm_cell_layer_call_and_return_conditional_losses, lstm_cell_1_layer_call_fn, lstm_cell_1_layer_call_and_return_conditional_losses while saving (showing 5 of 5). These functions will not be directly callable after loading.

INFO:tensorflow:Assets written to: ysp_models\assets

INFO:tensorflow:Assets written to: ysp_models\assets

39/39 - 11s - loss: 0.0145 - root_mean_squared_error: 0.1203 - val_loss: 0.0091 - val_root_mean_squared_error: 0.0955 - 11s/epoch - 285ms/step
Epoch 2/100

WARNING:absl:Found untraced functions such as _update_step_xla, lstm_cell_layer_call_fn, lstm_cell_layer_call_and_return_conditional_losses, lstm_cell_1_layer_call_fn, lstm_cell_1_layer_call_and_return_conditional_losses while saving (showing 5 of 5). These functions will not be directly callable after loading.

INFO:tensorflow:Assets written to: ysp_models\assets

INFO:tensorflow:Assets written to: ysp_models\assets

39/39 - 7s - loss: 0.0015 - root_mean_squared_error: 0.0381 - val_loss: 0.0019 - val_root_mean_squared_error: 0.0434 - 7s/epoch - 174ms/step
Epoch 3/100

WARNING:absl:Found untraced functions such as _update_step_xla, lstm_cell_layer_call_fn, lstm_cell_layer_call_and_return_conditional_losses, lstm_cell_1_layer_call_fn, lstm_cell_1_layer_call_and_return_conditional_losses while saving (showing 5 of 5). These functions will not be directly callable after loading.

model = load_model('ysp_models/')

predicted_stock_price = model.predict(X_test)

# invert predictions
predicted_stock_price = sc.inverse_transform(predicted_stock_price)
y_test = sc.inverse_transform([y_test])
y_test = y_test.reshape(-1, 1)

7/7 [==============================] - 1s 13ms/step

plt.plot(y_test, color = "red", label = "Real Stock Price")
plt.plot(predicted_stock_price, color = "blue", label = "Predicted Stock Price")
plt.title("Yahoo Stock Price Prediction")
plt.xlabel("Time")
plt.ylabel("Stock Price")
plt.legend()
plt.show()

mse = mean_squared_error(y_test, predicted_stock_price)
print(f"MSE === {mse}")

MSE === 1497.1658832307894

	Date	High	Low	Open	Close	Volume	Adj Close
Date
2015-11-23	2015-11-23	2095.610107	2081.389893	2089.409912	2086.590088	3.587980e+09	2086.590088
2015-11-24	2015-11-24	2094.120117	2070.290039	2084.419922	2089.139893	3.884930e+09	2089.139893
2015-11-25	2015-11-25	2093.000000	2086.300049	2089.300049	2088.870117	2.852940e+09	2088.870117

	High	Low	Open	Close	Volume	Adj Close
count	1825.000000	1825.000000	1825.000000	1825.000000	1.825000e+03	1825.000000
mean	2660.718673	2632.817580	2647.704751	2647.856284	3.869627e+09	2647.856284
std	409.680853	404.310068	407.169994	407.301177	1.087593e+09	407.301177
min	1847.000000	1810.099976	1833.400024	1829.079956	1.296540e+09	1829.079956
25%	2348.350098	2322.250000	2341.979980	2328.949951	3.257950e+09	2328.949951
50%	2696.250000	2667.840088	2685.489990	2683.340088	3.609740e+09	2683.340088
75%	2930.790039	2900.709961	2913.860107	2917.520020	4.142850e+09	2917.520020
max	3645.989990	3600.159912	3612.090088	3626.909912	9.044690e+09	3626.909912

	Open	Close	Adj Close	High	Low
Open	1.000000	0.998344	0.998344	0.999328	0.998794
Close	0.998344	1.000000	1.000000	0.998958	0.999020
Adj Close	0.998344	1.000000	1.000000	0.998958	0.999020
High	0.999328	0.998958	0.998958	1.000000	0.998154
Low	0.998794	0.999020	0.999020	0.998154	1.000000

Read data¶

Exploratory Data Analysis¶

Retrieve Opening stock prices¶

Scale the data¶

Split data into training, validation and test sets¶

Structure the dataset for LSTM model¶

Define the layers of the LSTM¶

Create the model and save the best model only¶

load the saved model¶

Mean Squared Error¶