import numpy as np import pandas as pd import matplotlib.pyplot as plt from mpl_toolkits.mplot3d import Axes3D from sklearn.model_selection import train_test_split import gradio as gr import plotly.graph_objects as go class LinearRegression: def __init__(self, eta = 0.01, n_iter = 1000): self.eta = eta self.n_iter = n_iter self.w = None self.b = None self.Lambda = 0.7 self.Min = None self.Max = None self.cost_history = None def ScaleData(self, X): return (X - self.Min) / (self.Max - self.Min); def compute_cost(self, X, y, w, b): m = len(y); J = np.sum(((np.dot(X,w) + b) - y) ** 2) / (2 * m); return J; def SubGrad(X, y, w, b): new_W = np.zeros((3,1)); m = X.shape[0]; A = X; dW = (((np.dot(X,w) + b) - y) / m).reshape(m,1); A = A*dW; new_W = np.sum(A, axis = 0).reshape((X.shape[1],1)); return new_W; def Grad_de(self,X, y, X_test, y_test, w, eta, b, lamda, decay_rate = 0.045): cost_history = []; new_W = w; Fake_W = np.zeros((X.shape[1],1)); S_corrected_w = np.zeros((X.shape[1],1)); new_B = 0; Fake_B = 0; S_corrected_b = 0; W_prev = Fake_W; m = len(X); iter = 1; count = 0; new_eta = eta; m = X.shape[0]; for iter in range(0,1000): new_W = self.SubGrad(X,y,w, lamda, b); new_B = 0; count = count + 1; new_B = np.sum(((np.dot(X,w) + b) - y) / m); Fake_W = 0.9 * Fake_W + 0.1 * new_W; Fake_B = 0.9 * Fake_B + 0.1 * new_B; S_corrected_w = 0.99 * S_corrected_w + 0.01 * (new_W ** 2); S_corrected_b = 0.99 * S_corrected_b + 0.01 * (new_B ** 2); w = w - (new_eta / (np.sqrt(S_corrected_w) + 1e-8)) * Fake_W; b = b - (new_eta / (np.sqrt(S_corrected_b) + 1e-8)) * Fake_B; cost_history.append(self.compute_cost(X_test,y_test,w,b)); new_eta = eta / (1 + np.floor(count/50) * decay_rate); return (w,b, cost_history); def standardize(self,X): self.Max = np.zeros(X.shape[1]); self.Min = np.zeros(X.shape[1]); for i in range(0,len(X)): self.Min[i] = 10000000; self.Max[i] = -10000000; for i in range(0,len(X)): for j in range(0,X.shape[1]): if self.Max[j] < X[i][j]: self.Max[j] = (X[i][j]); if self.Min[j] > X[i][j]: self.Min[j] = (X[i][j]); return (X - self.Min) / (self.Max - self.Min); def fit(self, X, y): self.w = (np.random.rand(X.shape[1])*0.01).reshape(X.shape[1],1); self.b = 0 self.cost = [] X = self.standardize(X); (self.w,self.b, self.cost_history) = self.Grad_de(X, y, self.w, self.b, self.eta, self.n_iter, self.Lambda); return self def predict(self, X): XPr = self.ScaleData(X, self.Min, self.Max); Xans = np.dot(XPr,self.w) + self.b; x1 = Xans[0][0]; return x1 df = pd.read_csv("C:\Project\Kaggle\Cali_housing_Price\housing_price_dataset.csv"); Data = pd.read_csv("C:\Project\Kaggle\Cali_housing_Price\housing_price_dataset.csv"); y = Data['Price']; X = Data.drop(columns= ['Price']); for i in range(X['Neighborhood'].shape[0]): if X.loc[i,'Neighborhood'] == 'Suburb': X.loc[i,'Neighborhood'] = 2; elif X.loc[i,'Neighborhood'] == 'Urban': X.loc[i,'Neighborhood'] = 3; else: X.loc[i,'Neighborhood'] = 1; Dx = X.to_numpy(dtype= np.float64); Dy = y.to_numpy(dtype= np.float64); X_train, X_test, y_train, y_test = train_test_split(Dx, Dy, test_size=0.02, random_state=42); w = (np.random.rand(X.shape[1])*0.01).reshape(X.shape[1],1); y_train = y_train.reshape(len(y_train),1); y_test = y_test.reshape(len(y_test),1); HousePriceModel = LinearRegression(); HousePriceModel.fit(X_train, y_train); def HousePrice(SquareFeet, Bedrooms, Bathrooms, Neighborhood, YearBuilt): NumNeighborhood = 0; if Neighborhood == 'Suburb': NumNeighborhood = 2; elif Neighborhood == 'Urban': NumNeighborhood = 3; elif Neighborhood == 'Rural': NumNeighborhood = 1; else: raise gr.Error("Invalid Neighborhood"); if YearBuilt > 2024 or YearBuilt < 1900: raise gr.Error("Invalid Year Built"); if SquareFeet < 0: raise gr.Error("Invalid Square Feet"); if Bedrooms < 0: raise gr.Error("Invalid Bedrooms"); if Bathrooms < 0: raise gr.Error("Invalid Bathrooms"); X = np.array([SquareFeet, Bedrooms, Bathrooms, NumNeighborhood, YearBuilt]).reshape(1,5); YPredict = HousePriceModel.predict(X); #### Filter the data filter_df = df[(df['SquareFeet'] >= SquareFeet - 20) & (df['SquareFeet'] <= SquareFeet + 20) & (df['Bedrooms'] >= Bedrooms - 0) & (df['Bedrooms'] <= Bedrooms + 0) & (df['Bathrooms'] >= Bathrooms - 0) & (df['Bathrooms'] <= Bathrooms + 0) & (df['YearBuilt'] >= YearBuilt - 5) & (df['YearBuilt'] <= YearBuilt + 5)]; df_list = filter_df.values.tolist() fig = go.Figure() fig.add_trace(go.Scatter( customdata=df_list, x = filter_df['SquareFeet'].tolist(), y = filter_df['Price'].tolist(), mode = 'markers', marker = dict(color = 'blue'), hoverinfo="text", hovertemplate= 'Square Feet: %{x}
Bedrooms: %{customdata[1]}
Bathrooms: %{customdata[2]}
Neighborhood: %{customdata[3]}
Year Built: %{customdata[4]}
Price: %{y}', name = 'House Price Actual' )) fig.add_trace(go.Scatter( x = [SquareFeet], y = [YPredict], mode = 'markers', marker = dict(color = 'red'), hovertext = 'Predicted Price', name = 'House Price Prediction' )) fig.update_layout() return YPredict, fig with gr.Blocks() as demo: gr.Markdown(""" # House Price Prediction This is a simple model to predict the price of a house based on its features. The database's feature have min sqaure feet is about 1000 and max is 3000. """) gr.Markdown(""" Enter the features of the house and click 'Predict Price' to see the predicted price. You can also click 'Filter Map' to see the actual prices of houses with similar features on a map. """) with gr.Column(): with gr.Row(): SquareFeet = gr.Number(value=250, label="Square Feet") Bedrooms = gr.Number(value=3, label="Bedrooms") Bathrooms = gr.Number(value=1, label="Bathrooms") Neighborhood = gr.Radio(["Suburb", "Urban", "Rural"], label="Neighborhood") YearBuilt = gr.Number(value=2020, label="Year Built") gr.Button("Predict Price").click( fn=HousePrice, inputs=[SquareFeet, Bedrooms, Bathrooms, Neighborhood, YearBuilt], outputs=[gr.Textbox(label="Predicted Price"), gr.Plot(label="Similar Houses")] ) demo.launch()