House Prices: Advanced Regression Techniques

# Import necessary libraries
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

# Load training and test datasets
train = pd.read_csv('homedata/train.csv')
test = pd.read_csv('homedata/test.csv')

# Print the unique data types in the training dataset
print(train.dtypes.unique())

# Select numerical and categorical columns from the training dataset
num_col = train.select_dtypes(exclude='object')
cat_col = train.select_dtypes(exclude=['int64', 'float64'])

# Display summary statistics for numerical columns
num_col.describe(include='all')

# Plot heatmap of missing values in numerical columns
plt.figure(figsize=(15,8))
sns.heatmap(num_col.isnull(), yticklabels=0,  cbar=False, cmap='viridis')

# Copy numerical columns to X and separate target variable y
X = num_col.copy()
y = X.pop('SalePrice')
X.isnull().sum()

# Display the number of missing values in each column of X
f, ax=plt.subplots(figsize=(20,2))
sns.heatmap(X.corr().iloc[8:9,:], annot=True, linewidths=.8, fmt='.1f', ax=ax)

# Plot kernel density estimate for 'MasVnrArea' column
sns.kdeplot(X.MasVnrArea,Label='MasVnrArea',color='g')

# Replace missing values in 'MasVnrArea' column with 0
X.MasVnrArea.replace(np.nan,0,inplace=True)

# Plot heatmap of correlations for another specific row in X
plt.figure(figsize=(20,2))
sns.heatmap(X.corr().iloc[25:26,:], annot=True, linewidths=.8, fmt='.1f')

# Fill missing values in 'GarageYrBlt' column with values from 'YearBuilt' column
X.GarageYrBlt.fillna(X.YearBuilt, inplace=True)
X[['GarageYrBlt','YearBuilt']]

# Plot heatmap of correlations for another specific row in X
plt.figure(figsize=(20,2))
sns.heatmap(X.corr().iloc[2:3,:], annot=True, linewidths=.8, fmt='.1f')

# Replace missing values in 'LotFrontage' column with the mean value
X.LotFrontage.replace(np.nan, X.LotFrontage.mean(), inplace=True)

# Import feature selection modules from sklearn
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection  import chi2

# Select top 30 features using chi-squared test
bestfeatures = SelectKBest(score_func=chi2, k=30)
fit = bestfeatures.fit(X,y)
# Create a DataFrame with feature scores
dfscores = pd.DataFrame(fit.scores_)
dfcolumns = pd.DataFrame(X.columns)

featurescores = pd.concat([dfcolumns,dfscores], axis=1)
featurescores.columns = ['Feature', 'Score']
featurescores.sort_values(by='Score', ascending=False)

# Import ExtraTreesClassifier from sklearn
from sklearn.ensemble import ExtraTreesClassifier

# Fit ExtraTreesClassifier model and plot feature importances
model = ExtraTreesClassifier()
model.fit(X,y)
feat_importance = pd.Series(model.feature_importances_, index=X.columns)
feat_importance.nlargest(30).plot(kind='barh', figsize=(30,10))

# Create a union of top features from both methods and remove 'Id' column
feats_tree = set(list(feat_importance.nlargest(30).index))
feats_chi = set(list(featurescores.Feature[:30]))
union_feat = feats_tree.union(feats_chi)
union_feat.remove('Id')
X = X[union_feat]
X.info()

# Import necessary modules for model training and evaluation
from sklearn.metrics import mean_absolute_error as mae 
from sklearn.model_selection import train_test_split as tt 
from sklearn.ensemble import RandomForestRegressor as rr 

# Split the data into training and validation sets
train_X, val_X, train_Y, val_Y = tt(X, y, random_state=23)

# Train RandomForestRegressor model and make predictions
forest_model = rr(random_state=12, max_depth=9, n_estimators=200)
forest_model.fit(X,y)
prediction = forest_model.predict(val_X)

# Calculate mean absolute error of the predictions
mae(val_Y,prediction)

# Select numerical columns from the test dataset
test_X = test.select_dtypes(exclude=['object'])

# Display the number of missing values in each column of test_X
X = num_col.copy()
test_X.isnull().sum()

# Replace missing values in specific columns of test_X
test_X.LotFrontage.replace(np.nan, test_X.LotFrontage.mean(), inplace=True)
test_X.MasVnrArea.replace(np.nan, test_X.MasVnrArea.mean(), inplace=True)
test_X.GarageYrBlt.fillna(X.YearBuilt, inplace=True)

# Define columns to plot and display their summary statistics
plot_these = ['BsmtFinSF1', 'BsmtFinSF2', 'BsmtUnfSF', 'TotalBsmtSF', 
            'BsmtFullBath', 'BsmtHalfBath', 'GarageCars', 'GarageArea']

print(test_X[plot_these].describe())

# Plot kernel density estimates for specified columns
plt.figure(figsize=(18,20))
for indexo, item in enumerate(plot_these):
    plt.subplot(4,2,indexo + 1)
    sns.kdeplot(test_X[item])

# Replace missing values in specified columns with median values
for item in plot_these:
    test_X[item].replace(np.nan, test_X[item].median(), inplace=True)

test_X.isnull().sum()

# Make predictions on the test dataset using the trained model
test_pred = forest_model.predict(test_X[union_feat])
test_pred 

# Create a DataFrame with the test predictions and save to CSV
output = pd.DataFrame({'Id': test.Id,
                      'SalePrice': test_pred})
output.to_csv('submission_numeric_20.csv', index=False)