OsmanBahadirYilmaz_Solution

HomeWork_oby.py

@@ -0,0 +1,161 @@
+import numpy as np
+import matplotlib.pyplot as plt
+import pandas as pd
+from sklearn.metrics import r2_score
+
+data=pd.read_csv('weatherAUS.csv')
+df = pd.read_csv('weatherAUS.csv')
+
+#dealing with nan values
+df = df.fillna(df.mean())
+
+#clear data
+selected_columns = ['Location', 'MinTemp', 'MaxTemp', 'Rainfall', 'WindGustDir', 'WindGustSpeed', 'WindSpeed9am',
+                    'WindSpeed3pm', 'Pressure9am', 'Pressure3pm', 'Humidity9am', 'Humidity3pm']
+
+X = df[selected_columns]
+
+
+#catagorical to numeric
+print(X['Location'].nunique())
+
+cat_columns=['Location','WindGustDir']
+for i in cat_columns:
+
+# Turn my column into a dummy value
+    dummy = pd.get_dummies(X[i])   
+    X = pd.concat([X, dummy], axis=1).drop(i, axis=1)
+
+
+y=X['MaxTemp']
+X.drop('MaxTemp', axis=1,inplace=True)
+
+
+from sklearn.model_selection import train_test_split,cross_val_score
+x_train, x_test,y_train,y_test = train_test_split(X,y,test_size=0.33, random_state=0)
+
+#1-Linear Regression
+
+from sklearn.linear_model import LinearRegression, Ridge , Lasso
+lin_reg = LinearRegression()
+lin_reg.fit(x_train,y_train)
+
+#scores
+y_hat=lin_reg.predict(x_test)
+from sklearn.metrics import mean_squared_error , mean_absolute_error
+
+print('--------------------')
+print('Linear regression')
+mae = mean_absolute_error(y_true= y_test , y_pred= y_hat)
+print('Mean Absolute Error (MAE) with test data' ,mae)
+
+mse = mean_squared_error(y_true=y_test , y_pred= y_hat)
+print("Mean Squared Error (MSE) with test data" , mse)
+
+
+
+#test with all values
+
+lin_reg2 = LinearRegression()
+lin_reg2.fit(X,y)
+
+y_hat=lin_reg2.predict(X)
+
+mae = mean_absolute_error(y_true= y , y_pred=y_hat )
+print('Mean Absolute Error (MAE) with all data' ,mae)
+
+mse = mean_squared_error(y_true=y , y_pred= y_hat)
+print("Mean Squared Error (MSE) with all data" , mse)
+
+print('R2 score with all data',r2_score(y, lin_reg.predict(X)))
+
+
+
+#2 Polyynomial Regression
+from sklearn.preprocessing import PolynomialFeatures
+poly_features = PolynomialFeatures(degree=2)
+
+X_poly = poly_features.fit_transform(X)
+
+poly_regression = LinearRegression()
+poly_regression.fit(X_poly, y)
+
+# Predicting values
+X_pred_poly = poly_features.transform(X)
+y_hat_poly = poly_regression.predict(X_pred_poly)
+print('--------------------')
+print('Polynomial Regrassion')
+mae = mean_absolute_error(y_true= y , y_pred=y_hat )
+print('Mean Absolute Error (MAE) with all data' ,mae)
+
+mse = mean_squared_error(y_true=y , y_pred= y_hat)
+print("Mean Squared Error (MSE) with all data" , mse)
+
+print('R2 score with all data',r2_score(y, lin_reg.predict(X)))
+
+#3 lasso and ridge
+lasso = Lasso()
+ridge = Ridge()
+lasso_fit = lasso.fit(x_train , y_train)
+ridge_fit = ridge.fit(x_train , y_train)
+
+print("RIDGE -> Score on Training Set ", ridge.score(x_train,y_train).round(4) , "\tScore on testing set ", ridge.score(x_test,y_test).round(4))
+print("LASSO -> Score on Training Set ", lasso.score(x_train,y_train).round(4) , "\tScore on testing set ", lasso.score(x_test,y_test).round(4))
+
+
+cross_val_score(ridge, x_train, y_train, cv = 10, scoring = 'r2')
+
+print("cross validation score MLR " , cross_val_score(lin_reg, x_train, y_train, cv = 10, scoring = 'r2').mean().round(6))
+print("cross validation score RIDGE " ,  cross_val_score(ridge, x_train, y_train, cv = 10, scoring = 'r2').mean().round(6))
+print("cross validation score LASSO " , cross_val_score(lasso, x_train, y_train, cv = 10, scoring = 'r2').mean().round(6))
+
+
+#Scaling data
+from sklearn.preprocessing import MinMaxScaler
+scaler = MinMaxScaler(feature_range=(0,1))
+scaler_y = MinMaxScaler(feature_range=(0,1))
+X_normalized = scaler.fit_transform(X)
+Y_normalized = scaler_y.fit_transform(np.array(y).reshape(len(y),1))
+
+X_train_n, X_test_n, y_train_n, y_test_n = train_test_split(X_normalized, Y_normalized, test_size=0.2, shuffle = True, random_state = 15)
+
+mlr_n = LinearRegression()
+ridge_n = Ridge()
+lasso_n = Lasso()
+
+mlr_n_fit = mlr_n.fit(X_train_n , y_train_n)
+ridge_n_fit = ridge_n.fit(X_train_n , y_train_n)
+lasso_n_fit = lasso_n.fit(X_train_n , y_train_n)
+
+print("MLR -> Score on Normalized Training Set ", mlr_n.score(X_train_n,y_train_n).round(4) , "\tScore on Normalized testing set ", mlr_n.score(X_test_n,y_test_n).round(4))
+print("RIDGE -> Score on Normalized Training Set ", ridge_n.score(X_train_n,y_train_n).round(4) , "\tScore on Normalized testing set ", ridge_n.score(X_test_n,y_test_n).round(4))
+print("LASSO -> Score on Normalized Training Set ", lasso_n.score(X_train_n,y_train_n).round(4) , "\tScore on testing set ", lasso_n.score(X_test_n,y_test_n).round(4))
+
+print("cross validation score MLR (normalized) " , cross_val_score(mlr_n, X_train_n, y_train_n, cv = 10, scoring = 'r2').mean().round(4))
+print("cross validation score RIDGE (normalized) " ,  cross_val_score(ridge_n, X_train_n, y_train_n, cv = 10, scoring = 'r2').mean().round(4))
+print("cross validation score LASSO (normalized) " , cross_val_score(lasso_n, X_train_n, y_train_n, cv = 10, scoring = 'r2').mean().round(4))
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

Results.txt

@@ -0,0 +1,28 @@
+
+Mean Absolute Error (MAE)  2.0793131035472134
+Mean Squared Error (MSE)  7.324441316105051
+
+Mean Absolute Error (MAE) with all data 2.0771314409276798
+Mean Squared Error (MSE) with all data 7.313528794189804
+
+Linear R2 score 0.8543937666026298
+
+RIDGE -> Score on Training Set  0.8544 	Score on testing set  0.8544
+LASSO -> Score on Training Set  0.814 	Score on testing set  0.8146
+
+cross_val_score(ridge, x_train, y_train, cv = 10, scoring = 'r2')
+Out[117]: 
+array([0.85456572, 0.850983  , 0.85393335, 0.85654553, 0.86212046,
+       0.84730399, 0.84921717, 0.85548548, 0.85526248, 0.85529536])
+
+cross validation score MLR  0.854071
+cross validation score RIDGE  0.854071
+cross validation score LASSO  0.813897
+
+MLR -> Score on Normalized Training Set  0.8544 	Score on Normalized testing set  0.8543
+RIDGE -> Score on Normalized Training Set  0.8544 	Score on Normalized testing set  0.8543
+LASSO -> Score on Normalized Training Set  0.0 	Score on testing set  -0.0
+
+cross validation score MLR (normalized)  0.8542
+cross validation score RIDGE (normalized)  0.8542
+cross validation score LASSO (normalized)  -0.0