Added .py script

Stock Price Prediction Using Python & LSTM (Long Short Term Memory).py

@@ -0,0 +1,158 @@
+#!/usr/bin/env python
+# coding: utf-8
+
+# In[284]:
+
+
+#IMPORTING LIBRARIES 
+import math
+import numpy as np
+import pandas as pd 
+import pandas_datareader
+import matplotlib.pyplot as pyplot
+
+#A sequential model takes into account the order of things happening and is linear, it is also fairly simple to use.
+from keras.models import Sequential
+
+#Dense layer is the regular deeply connected neural network layer. Essentially a vanilla NN
+#The Dropout layer randomly sets input units to 0 with a frequency of rate at each step during training time, which helps prevent overfitting. Input data may have some of the unwanted data, usually called as Noise. Dropout will try to remove the noise data and thus prevent the model from over-fitting.
+#LSTM is a Recurrent neural network that helps in sequential data (Data that needs to preserve order) and also helps fight the forgetting problem in vanilla RNN 
+from keras.layers import LSTM,Dropout,Dense
+
+#Transforms any actual range into a range of 0 to 1 so it is easier to work with, essentially the Sigmoid function / ReLU
+from sklearn.preprocessing import MinMaxScaler
+
+
+# In[343]:
+
+
+#IMPORTING DATA
+df = pandas_datareader.DataReader('AAPL',data_source='yahoo', start= "2012-01-01", end="2019-12-31" )
+df
+
+
+# In[344]:
+
+
+#VISUALISING CLOSING PRICES
+
+#using my prefered styling for our graphs
+pyplot.style.use('dark_background')
+#Plotting the graph
+pyplot.figure(figsize=(20,20))
+pyplot.plot(df['Close'])
+pyplot.title("Closing Prices History from 2012-2019")
+pyplot.xlabel('Date', fontsize=20)
+pyplot.ylabel('Close Price USD ($)', fontsize=20)
+pyplot.show()
+
+
+# In[345]:
+
+
+data = df.filter(['Close'])
+#convert data frame to be numpy compatible 
+dataset = data.values
+
+#The rest is used to test 
+train_len = math.ceil(len(dataset) * 0.8)
+
+
+# In[346]:
+
+
+#Scaling the data and Creating training data sets 
+#Converting the data from arrays of hundreds to arrays of 0 to 1s so we can work with them efficiently 
+scaler = MinMaxScaler(feature_range=(0,1))
+scaled_data = scaler.fit_transform(dataset)
+#Creating the training data set 
+train_data = scaled_data[0:train_len, :]
+
+#x is the independant variable and y is dependant
+x_train,y_train=[],[]
+
+for i in range(60,len(train_data)):
+    #We are leaving out some of the data set to test our model later
+    x_train.append(scaled_data[i-60:i,0])
+    y_train.append(scaled_data[i,0])
+
+#Converting to numpy arrays / tensors to be used in models
+x_train,y_train=np.array(x_train),np.array(y_train)
+#We Reshape the model as LSTM requires a 3d shape so we just insert 1 to the 3rd dimension
+x_train=np.reshape(x_train,(x_train.shape[0],x_train.shape[1],1))
+
+
+# In[347]:
+
+
+#Build the LSTM model 
+model = Sequential()
+model.add(LSTM(50, return_sequences=True, input_shape = (x_train.shape[1],1)))
+model.add(LSTM(50,return_sequences=False))
+model.add(Dense(25))
+model.add(Dense(1))
+
+#Compilig the Model 
+#A loss / cost function tells the computer how wrong it was from the actual result and a optimizer acts as a catalyst
+model.compile(loss='mean_squared_error',optimizer='adam')
+
+#Training the model
+model.fit(x_train,y_train,batch_size=1,epochs=1)
+#batch_size = how many batches at a time 
+#epochs = how many times to run this program
+
+
+# In[348]:
+
+
+#Creating Testing Data
+test_data = scaled_data[train_len - 60: , :]
+#Create data sets 
+test_set = []
+actual_values = dataset[train_len:,:]
+
+#Using Rest of the dataframe
+for i in range(60,len(test_data)):
+    test_set.append(test_data[i-60:i,0])
+
+#Converting test_set to numpy array and reshaping it
+test_set = np.array(test_set)
+test_set = np.reshape(test_set,(test_set.shape[0],test_set.shape[1],1))
+
+
+# In[349]:
+
+
+#Get predicted prices
+predicted_closing_price = model.predict(test_set)
+predicted_closing_price = scaler.inverse_transform(predicted_closing_price)
+
+
+# In[350]:
+
+
+#Getting standard deviation / RMSE (Root Mean Squared Error) it is a way to calculate the accuracy of our model
+rmse =np.sqrt(np.mean(((predicted_closing_price- actual_values)**2)))
+rmse = np.sqrt(np.mean(np.power((np.array(actual_values)-np.array(predicted_closing_price)),2)))
+rmse = np.sqrt(((predicted_closing_price - actual_values) ** 2).mean())
+rmse =  round(rmse,2)
+
+
+# In[352]:
+
+
+#Plot the data 
+train = data[:train_len]
+valid = data[train_len:]
+valid['Predictions'] = predicted_closing_price
+#Visualisation
+pyplot.figure(figsize=(20,20))
+pyplot.title("Closing Prices History from 2012-2019")
+pyplot.xlabel('Date', fontsize=20)
+pyplot.ylabel('Close Price USD ($)', fontsize=20)
+pyplot.plot(train['Close'])
+pyplot.plot(valid[['Close','Predictions']])
+pyplot.legend(['Training Set','Actual Values','Predictions'],loc="lower right")
+pyplot.text(15250,75,f'RMSE of given prediction is {rmse}', fontsize = 22)
+pyplot.show()
+