lab5b4

'''
Ritam Chakraborty
B20127
7439257709
'''

# import libraries
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from sklearn.preprocessing import PolynomialFeatures
from sklearn.linear_model import LinearRegression

# function for rmse
def rmse(pred, actual):
    return np.sqrt(np.sum(np.square(pred-actual))/len(pred))

# import the datasets
df_train = pd.read_csv("abalone-train.csv")
df_test = pd.read_csv("abalone-test.csv")
corrs = df_train.corr()["Rings"]

# take out the input and target variables from the dataset
P = [2, 3, 4, 5]
X = pd.DataFrame(df_train[df_train.keys()[:-1]])
Y = df_train["Rings"]
X_test = pd.DataFrame(df_test[df_test.keys()[:-1]])
Y_test = df_test["Rings"]

RMSE = []
for p in P:
    # transform each of the input vectors for polynomial regression
    polyFeat = PolynomialFeatures(p)
    poly_inp = polyFeat.fit_transform(X)
    # perform linear regression on the transformed input vectors to perform polynomial regression
    LinReg = LinearRegression()
    LinReg.fit(poly_inp, Y)
    # predict for the training sample and compute rmse with them
    train_pred = LinReg.predict(poly_inp)
    print("RMSE in training dataset when p =", str(p)+":", rmse(train_pred, Y))
    print("Regression accuracy =", 100-len(Y)*rmse(train_pred, Y)/np.sum(Y)*100)
    RMSE.append(rmse(train_pred, Y))

# bar plot of rmse with values p = 2, 3, 4, 5
plt.bar(P, RMSE)
plt.title("RMSE vs degree of Polynomial")
plt.show()

RMSE = []
for p in P:
    # convert into polynomial form
    polyFeat = PolynomialFeatures(p)
    poly_inp = polyFeat.fit_transform(X)
    # perform linear regression on transformed vectors to change into polynomial regression
    LinReg = LinearRegression()
    LinReg.fit(poly_inp, Y)
    # predict for the test dataset and compute rmse
    test_poly = polyFeat.fit_transform(X_test)
    test_pred = LinReg.predict(test_poly)
    print("RMSE in test dataset when p =", str(p)+":", rmse(test_pred, Y_test))
    print("Regression accuracy =", 100-len(Y_test)*rmse(test_pred, Y_test)/np.sum(Y_test)*100)
    RMSE.append(rmse(test_pred, Y_test))

# bar plot of rmse with values p = 2, 3, 4, 5
plt.bar(P, RMSE)
plt.title("RMSE vs degree of Polynomial")
plt.show()

# find the best fitting polynomial curve and apply regression on it
best_fit = P[np.argmin(RMSE)]
print("Best fit is with p =", best_fit)
polyFeat = PolynomialFeatures(best_fit)
regressor = LinearRegression().fit(polyFeat.fit_transform(X), Y)

# Scatter plot of actual vs predicted values
plt.scatter(Y_test, regressor.predict(polyFeat.fit_transform(X_test)), alpha = 0.5)
plt.xlabel("Actual")
plt.ylabel("Predicted")
plt.title("Predicted vs Actual")
plt.show()