diff --git a/api/run.py b/api/run.py
index 1925bd37f46a2b4a5a9c14ca3c0f660570afca2d..1de262669dbe705b5eee8d383d9ccf1eb13b1248 100644
--- a/api/run.py
+++ b/api/run.py
@@ -1,7 +1,13 @@
+#====================================================
+# Written By Don Gunasinha =
+# Master in Data Science =
+#====================================================
+
+# libarary declare
from app import app
import pandas as pd
-import datetime as dt
+import datetime as dt
import numpy as np
from sqlalchemy import create_engine
@@ -11,196 +17,257 @@ from tensorflow import keras
from keras.models import Sequential
from keras.layers import Dense, LSTM, Dropout
+from flask import Flask, jsonify, request, make_response
+from flask import flash, request
+
+from datetime import datetime, timedelta
import matplotlib.pyplot as plt
+#=================================================
+
# Globally Declare data set
+# Data Frames
+raw_data_set=pd.DataFrame()
data_set =pd.DataFrame()
-
+index_data_set =pd.DataFrame()
tranning_set=pd.DataFrame()
+result=pd.DataFrame()
-test_result=pd.DataFrame()
-
+# numpy Arrays
data_arr_x=np.array([])
data_arr_y=np.array([])
-
x_train = np.array([])
y_train = np.array([])
x_test = np.array([])
y_test = np.array([])
+predictions = np.array([])
-test_predict = np.array([])
+# Lstm Model
+model = Sequential()
+feature_length = 100
-train_ind =0
+# Scaler
+scaler = MinMaxScaler(feature_range=(0,1))
-scaler = MinMaxScaler()
+#============================================
+# Support Functions definitions
-num_steps = 60
+
+# Read data from data base and re arrange data
def read_data_set():
# Declare global variable
global data_set
- global tranning_data_set
-
+ global raw_data_set
+ global index_data_set
# Define database connection
db_connection_str = 'mysql+pymysql://root:@localhost/csct'
db_connection = create_engine(db_connection_str)
# Read data in to Data Frame
- data_set = pd.read_sql('SELECT * FROM product__demnd', con=db_connection)
- # Validate Date
- data_set['Date']= pd.to_datetime(data_set['Date']).dt.date
- data_set.sort_values('Date', inplace=True)
- data_set['Date']=data_set['Date'].astype(str)
-
-
-
- # Remove Nan values
-
-
- # Clean Order Demand
-
- data_set['Order_Demand'] = data_set['Order_Demand'].str.replace('(',"")
- data_set['Order_Demand'] = data_set['Order_Demand'].str.replace(')',"")
- data_set['Order_Demand'] = data_set['Order_Demand'].astype('int64')
-
- tranning_data_set = data_set.groupby('Date')['Order_Demand'].sum().reset_index()
- # Reset index
- data_set=data_set.set_index(data_set['Date'])
- data_set.dropna(inplace=True)
-
- return data_set
-
-
-def reshape():
- global data_set
- global data_arr_x
- global data_arr_y
- global num_steps
- global tranning_data_set
- global scaler
+ raw_data_set = pd.read_sql('SELECT * FROM aw_product_demand', con=db_connection)
- items=tranning_data_set.filter(['Order_Demand'])
- item_arr=np.array(items.values)
- scaled_data = scaler.fit_transform(item_arr)
- print(scaled_data)
+ # Validate Date
+ raw_data_set['Date']= pd.to_datetime(raw_data_set['Date']).dt.date
+ raw_data_set['Order_Demand'] = raw_data_set['Order_Demand'].astype('int64')
+ #combine to single date
+ data_set = raw_data_set.groupby('Date')['Order_Demand'].sum().reset_index()
- data_arr_x, data_arr_y = lstm_data_transform(scaled_data,scaled_data, num_steps=num_steps)
- print ("The new shape of x is", data_arr_x.shape)
-
-
-
- original_data=scaler.inverse_transform(scaled_data)
- print(original_data)
-
-def split_data():
- global data_arr_x
- global data_arr_y
- global num_steps
-
- global x_train
- global y_train
+ data_set.sort_values('Date', inplace=True)
+ data_set['Date']=data_set['Date'].astype(str)
- global x_test
- global y_test
+ # Create index data frame
+ index_data_set=data_set
- global train_ind
+ index_data_set=index_data_set.set_index(index_data_set['Date'])
- train_ind = int(0.8 * len(data_arr_x))
- x_train = data_arr_x[:train_ind]
- y_train = data_arr_y[:train_ind]
- x_test = data_arr_x[train_ind:]
- y_test = data_arr_y[train_ind:]
+ return data_set
-def lstmmodel():
+# Function to create x and y data
+def Create_Features_and_Targets(data, feature_length):
+ # Declare Xand Y List
+ X = list()
+ Y = list()
+ for i in range(len(data) - feature_length -1):
+ # create X array shift by feature length
+ X.append(data[i:(i + feature_length), 0])
+
+ Y.append(data[i + feature_length, 0])
+ # convert to numpy array format
+ X = np.array(X)
+ Y = np.array(Y)
+ return X,Y
+
+# function to calculate prediction
+
+def predict_given_date(data, date, feature_length):
+ if date not in data.index:
+ data.loc[date]=0
+ idx = data.index.get_loc(date)
+ close_col = data.iloc[:,1:2]
+ close_col = close_col.iloc[idx - feature_length : idx,:].values
+ close_col = np.expand_dims(scaler.transform(close_col) , axis = 0)
+ Prediction = model.predict(close_col)
+ Prediction=np.array(Prediction).reshape(-1, 1)
+ Prediction = scaler.inverse_transform(Prediction)
+ return Prediction
+# =========================== End of Function =================
+
+#============================= Main funtion ===================
+def setup():
+
+ # Declare Global variables =================
+ global data_set
+ global predictions
+ global result
+
+ # Read Data from DataSet
+ read_data_set()
+
+ # Filter order Deamands array
+ data_set["Order_Demand"]=pd.to_numeric(data_set.Order_Demand,errors='coerce')
+ data = data_set.iloc[:,1:2]
+ data = data.values
+
+
+ data = scaler.fit_transform(data)
+
+
+ X_train,y_train= Create_Features_and_Targets(data,feature_length)
+ X_train = np.reshape(X_train,(X_train.shape[0],X_train.shape[1],1))
+
+ print(X_train)
+
+ # model
+ model = Sequential([
+ LSTM(100,return_sequences=True,input_shape=(X_train.shape[1],1)),
+ Dropout(0.3),
+ LSTM(100, return_sequences = False),
+ Dropout(0.3),
+
+ Dense(1),
+ ])
+ model.compile(optimizer='adam',loss="mean_squared_error")
+ model.summary()
+
+ # Training the model
+ history = model.fit(
+ X_train,
+ y_train,
+ epochs = 5,
+ batch_size = 12,
+ verbose=1,
+ )
+
+ testData = data_set.iloc[:,1:2] # Get 'Close' feature
+ y_real=testData.iloc[feature_length+1:,0:].values #Actual values
+ x_test = testData.iloc[:,0:].values # data to test
+ # normalizing the Data using Scaler.transform function
+ x_test = scaler.transform(x_test)
+ x_test, y_test = Create_Features_and_Targets(x_test, feature_length)
+ # Making data 3 dimensional
+ x_test = np.reshape(x_test,(x_test.shape[0],x_test.shape[1],1))
+ y_pred = model.predict(x_test)
+ predictions=np.array(scaler.inverse_transform(y_pred)).ravel()
+ result["Date"]=data_set.iloc[feature_length+1:]["Date"]
+ result["Predictions"]=predictions
+ result["OrderDemand"]=data_set.iloc[feature_length+1:]["Order_Demand"]
+
+ result.plot()
+ plt.show()
+ print(result)
- global test_predict
-
- model = Sequential()
-
- model.add(LSTM(512, activation='relu', input_shape=(num_steps, 1),
- return_sequences=False))
- model.add(Dense(units=256, activation='relu'))
+ #save model
+ model.save('lstm_model.h5')
+# ====================================================================
+# main function call
- model.add(Dense(units=1, activation='relu'))
-
-
- model.compile(optimizer="Adam", loss="mean_squared_error", metrics=['mae'])
- model.fit(x_train, y_train, epochs=25)
+setup()
- test_predict = model.predict(x_test)
-def plot_result():
+#====================================================================
- plt.style.use('ggplot')
- plt.figure(figsize=(20, 7))
- plt.plot(y_test, label="True value")
- plt.plot(test_predict.ravel(), label="Predicted value")
- plt.legend()
- plt.show()
+# end points declaration
+@app.route("/validation",methods=['GET'])
+def validation():
+ global result
+ result=result.set_index(result['Date'])
+ return result.to_json(orient='records')
-def lstm_data_transform(x_data, y_data, num_steps=5):
- """ Changes data to the format for LSTM training
-for sliding window approach """
- # Prepare the list for the transformed data
- X, y = list(), list()
- # Loop of the entire data set
- for i in range(x_data.shape[0]):
- # compute a new (sliding window) index
- end_ix = i + num_steps
- # if index is larger than the size of the dataset, we stop
- if end_ix >= x_data.shape[0]:
- break
- # Get a sequence of data for x
- seq_X = x_data[i:end_ix]
- # Get only the last element of the sequency for y
- seq_y = y_data[end_ix]
- # Append the list with sequencies
- X.append(seq_X)
- y.append(seq_y)
- # Make final arrays
- x_array = np.array(X)
- y_array = np.array(y)
- return x_array, y_array
-
-@app.route('/start',methods=['GET'])
-def start():
+@app.route("/forecast_to_date",methods=["POST"])
+def forecast():
global data_set
+ global scaler
+ global index_data_set
+ global model
- print(data_set.info())
+ # read incomming json data
+ data=request.get_json()
+ date=data['date']
+ new_date=datetime.strptime(date, "%Y-%m-%d").date()
+ new_date=new_date - timedelta(days=(feature_length-1))
+ new_date=new_date.strftime("%Y-%m-%d")
+ print(new_date)
- return data_set.to_json(orient='records')
-@app.route("/result_tranning",methods=['GET'])
-def result_tranning():
- global test_result
+ result=predict_given_date(index_data_set,new_date, feature_length)
+ df=pd.DataFrame()
+ # df=pd.DataFrame(data=result,columns=["Prediction"])
+ df['Date']=pd.date_range(start=new_date,periods=feature_length)
+ df=df.loc[::-1]
+ df['Prediction']=result
- test_len=len(tranning_data_set)-len(test_predict)
- test_result=tranning_data_set[test_len:]
- test_result['Predictions']=test_predict
- test_result['OrderDemand']=y_test
- # test_result=test_result.set_index(test_result['Date'])
- return test_result.to_json(orient='records')
+ df['Date']= pd.to_datetime(df['Date']).dt.date
+ df.sort_values('Date', inplace=True)
+ df['Date']=df['Date'].astype(str)
+ df=df.set_index(df['Date'])
+ df=df.tail(1)
+ return df.to_json(orient='records')
+@app.route("/forecast_to_range",methods=["POST"])
+def forecast_to_range():
+ global data_set
+ global scaler
+ global index_data_set
+ global model
+
+ index_data_set=data_set
+
+ index_data_set=index_data_set.set_index(index_data_set['Date'])
-read_data_set()
-reshape()
-split_data()
-lstmmodel()
-plot_result()
+ # read incomming json data
+ data=request.get_json()
+ date=data['date']
+ new_date=datetime.strptime(date, "%Y-%m-%d").date()
+ new_date=new_date - timedelta(days=(feature_length-1))
+ new_date=new_date.strftime("%Y-%m-%d")
+
+ result=predict_given_date(index_data_set,new_date, feature_length)
+ df=pd.DataFrame()
+ # df=pd.DataFrame(data=result,columns=["Prediction"])
+ df['Date']=pd.date_range(start=new_date,periods=feature_length)
+ df=df.loc[::-1]
+ df['Prediction']=result
+ df['Date']= pd.to_datetime(df['Date']).dt.date
+ df.sort_values('Date', inplace=True)
+ df['Date']=df['Date'].astype(str)
+ df=df.set_index(df['Date'])
+ return df.to_json(orient='records')
+
if __name__ == "__main__":
app.run()
\ No newline at end of file