import pandas as pandas
import numpy as numpy
import matplotlib.pyplot as plt




# punkt 4
def initialize_weights():
    # print("----------------------initialize_weights----------------------------")
    matrix = numpy.random.rand(x.shape[1] + 1, 1)
    return matrix

# punkt 5
def insert_coeficient_to_weights_matrix(x):
    # print("-----------------------insert_coeficient_to_weights_matrix---------------------")
    return numpy.insert(x, 0, 1, axis=1)

# punkt 6
def predict(x, macierzWspolRegresji):
    # print("----------------------------------predict-----------------------------------------")
    predictions = numpy.dot(x, macierzWspolRegresji)
    return predictions

# punkt 7a
def calculate_residual(predictions, df):
    # print("-------------------------------calculate_residual---------------------------------------")
    residuals = numpy.subtract(predictions, df)
    return residuals

#punkt 7b
def multiply_residual_by_features_matrix(x, df):
    # print("-------------------------------multiply_residual_by_features_matrix----------------------------------------")
    gradient_vector = numpy.dot(numpy.transpose(x), calc_residuals(prediction, df))
    return gradient_vector

#punkt 7
def calculate_gradient(predictions, df, x):
    # print("-------------------------------calculate_gradient----------------------------------------")
    residuals = calculate_residual(predictions, df)
    gradient_vector = multiply_residual_by_features_matrix(x, df)
    n = 35
    gradient_vector = gradient_vector * 1/n
    return gradient_vector

#punkt 8
def update_weights(macierzWspolRegresji, gradient_vector):
    # print("-------------------------------update_weights----------------------------------------")
    alfa = 0.0001
    matrix = numpy.subtract(macierzWspolRegresji, (alfa * gradient_vector))
    return matrix


def learn_and_fit(x, y):
    # print("-------------------------------learn and fit----------------------------------------")
    weights = initialize_weights()
    # print(weights)
    expanded_x = insert_coeficient_to_weights_matrix(x)
    # print(expanded_x)

    # punkt 9
    for i in range(140000):
        predictions = predict(expanded_x, weights)
        # print(predictions)
        gradient_vector = calculate_gradient(predictions, df, expanded_x)
        # print(gradient_vector)
        weights = update_weights(weights, gradient_vector)
    return weights


# ---------------------------------------------------------------------------------------
df = pandas.read_csv('Salary.csv')
x = df.iloc[:,:-2].values
y = df.iloc[:,-1].values.reshape(df.shape[0], 1)

obliczoneWspolRegresji = learn_and_fit(x, y)
x1 = obliczoneWspolRegresji.iloc[:,:-2].values
y1 = obliczoneWspolRegresji.iloc[:,-1].values.reshape(df.shape[0], 1)

# punkt 11
plt.plot(range(35), y, color="green")
plt.plot(range(35), y1*100, color="red")
plt.grid(True);
plt.show()