lstm and optimize

model.py

@@ -108,4 +108,206 @@ plt.show()
 # Monthly pivot table
 df = (df.pivot(index='Year', columns='Month', values='OrderDemand'))
 df = df.loc[:, ['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec']]
-df
+print(df)
\ No newline at end of file
+
+
+
+custom_dict = {'Jan':0, 'Feb':1, 'Mar':2, 'Apr':3, 'May':4, 'Jun':5,
+               'Jul':6, 'Aug':7, 'Sep':8, 'Oct':9, 'Nov':10, 'Dec':11}
+temp_data = data.copy()
+temp_data.Month.replace([1,2,3,4,5,6,7,8,9,10,11,12], ['Jan', 'Feb', 'Mar', 'Apr', 'May', 
+                                                               'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec'], inplace=True)
+df = temp_data[["OrderDemand", 'Month', 'Year']].groupby(["Year",
+                                                  "Month"]).sum().reset_index().sort_values(by=['Year', 
+                                                                                                 'Month'], ascending=True)
+df = df.iloc[df['Month'].map(custom_dict).argsort()]
+f, ax=plt.subplots(figsize=(15, 5))
+sns.pointplot(x='Month', y="OrderDemand", data=df, hue="Year")
+plt.show()
+
+print(df.describe())
+
+
+
+df = data[["OrderDemand", 'Year', 'Warehouse']].groupby(["Year",
+                                                         "Warehouse"]).sum().reset_index().sort_values(by=['Warehouse','Year'], ascending=False)
+f, ax=plt.subplots(figsize=(15, 5))
+sns.pointplot(x='Year', y="OrderDemand", data=df, hue="Warehouse")
+plt.show()
+
+df = (df.pivot(index='Year', columns='Warehouse', values='OrderDemand'))
+print(df)
+
+
+print(df.describe())
+
+
+#Product Category Based Analysis
+
+
+df = data[["OrderDemand",
+           'ProductCategory', 'Warehouse']].groupby(["ProductCategory",
+                                                     "Warehouse"]).sum().reset_index().sort_values(by=['OrderDemand'],
+                                                                                                   ascending=False)
+df = df.pivot(index='ProductCategory', columns='Warehouse', values='OrderDemand')
+print(df)
+
+#Forecast the Order Demand with LSTM Model
+
+df = data[(data['Date']>='2012-01-01') & (data['Date']<='2016-12-31')].sort_values('Date', ascending=True)
+df = df.groupby('Date')['OrderDemand'].sum().reset_index()
+print(df)
+
+# Visualize the order demand as time series
+plt.figure(figsize=(16, 8))
+plt.title("Order Demand Graph")
+plt.plot(df["Date"], df["OrderDemand"])
+plt.xlabel("Time", fontsize=14,)
+plt.ylabel("Order Demand", fontsize=14)
+plt.show()
+
+
+# Create new data with only the "OrderDemand" column
+orderD = df.filter(["OrderDemand"])
+# Convert the dataframe to a np array
+orderD_array = orderD.values
+# See the train data len
+train_close_len = math.ceil(len(orderD_array) * 0.8)
+print(train_close_len)
+
+
+# Normalize the data
+scaler = MinMaxScaler()
+scaled_data = scaler.fit_transform(orderD_array)
+print(scaled_data)
+
+
+
+# Create the training dataset
+train_data = scaled_data[0 : train_close_len, :]
+# Create X_train and y_train
+X_train = []
+y_train = []
+for i in range(60, len(train_data)):
+    X_train.append(train_data[i - 60 : i, 0])
+    y_train.append(train_data[i, 0])
+    if i <= 60:
+        print(X_train)
+        print(y_train)
+
+#  make X_train and y_train np array
+X_train, y_train = np.array(X_train), np.array(y_train)
+
+# reshape the data
+X_train = np.reshape(X_train, (X_train.shape[0], X_train.shape[1], 1))
+print(X_train.shape)
+
+
+# create the testing dataset
+test_data = scaled_data[train_close_len - 60 : , :]
+# create X_test and y_test
+X_test = []
+y_test = df.iloc[train_close_len : , :]
+for i in range(60, len(test_data)):
+    X_test.append(test_data[i - 60 : i, 0])
+
+
+# convert the test data to a np array and reshape the test data
+X_test = np.array(X_test)
+X_test = np.reshape(X_test, (X_test.shape[0], X_test.shape[1], 1))
+
+
+
+# Build the LSTM Model
+model = Sequential()
+
+model.add(LSTM(units=512, return_sequences=True, activation='relu', input_shape=(X_train.shape[1], 1)))
+
+
+model.add(LSTM(units=256, activation='relu', return_sequences=False))
+
+
+model.add(Dense(units=1))
+
+
+# compile the LSTM model
+model.compile(optimizer="Adam", loss="mean_squared_error", metrics=['mae'])
+
+
+
+# train the LSTM model
+model.fit(X_train, y_train,
+          epochs=3,
+          batch_size=100, 
+          verbose=1)
+
+
+
+# predict with LSTM model
+predictions = model.predict(X_test)
+predictions = scaler.inverse_transform(predictions)
+
+# plot the data
+train = orderD[:train_close_len]
+valid = orderD[train_close_len:]
+valid["Predictions"] = predictions
+#visualize the data
+plt.figure(figsize=(16, 8))
+plt.title("Forecast with LSTM Model")
+plt.xlabel("Time", fontsize=14)
+plt.ylabel("Order Demand", fontsize=14)
+plt.plot(df["Date"][:train_close_len], train["OrderDemand"])
+plt.plot(df["Date"][train_close_len:], valid[["OrderDemand", "Predictions"]])
+plt.legend(["Train", "Validation", "Predictions"], loc="lower right")
+plt.show()
+
+
+# change the parameters of first LSTM model and build the Optimized LSTM Model
+optimized_model = Sequential()
+
+optimized_model.add(LSTM(512, activation='relu', return_sequences=True, input_shape=(X_train.shape[1], 1)))
+
+optimized_model.add(LSTM(256, activation='relu', return_sequences=False))
+
+optimized_model.add(Dense(128))
+
+optimized_model.add(Dense(64))
+
+optimized_model.add(Dense(32))
+
+optimized_model.add(Dense(1))
+
+
+# compile the model
+optimized_model.compile(optimizer="Adam", loss="mean_squared_error", metrics=['mae'])
+
+
+
+# train the optimized model
+optimized_model.fit(X_train, y_train, 
+          batch_size=32, 
+          epochs=20, 
+          verbose=1)
+
+
+
+
+# Predict with optimized LSTM model
+o_predictions = optimized_model.predict(X_test)
+o_predictions = scaler.inverse_transform(o_predictions)
+
+# plot the data
+train = orderD[:train_close_len]
+valid = orderD[train_close_len:]
+valid["Predictions"] = o_predictions
+#visualize the data
+plt.figure(figsize=(16, 8))
+plt.title("Forecast with Optimized LSTM Model")
+plt.xlabel("Time", fontsize=14)
+plt.ylabel("Order Demand", fontsize=14)
+plt.plot(df["Date"][:train_close_len], train["OrderDemand"])
+plt.plot(df["Date"][train_close_len:], valid[["OrderDemand", "Predictions"]])
+plt.legend(["Train", "Validation", "Predictions"], loc="upper right")
+plt.show()
+
+