05-cali_ex.ipynb

Kernel: Python 3

California Housing

%matplotlib notebook
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from sklearn.linear_model import LinearRegression, Lasso, Ridge, ElasticNet
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import train_test_split
import seaborn as sns
from pandas.plotting import scatter_matrix

Load the Data: Visualize It

cali = pd.read_csv('data/cali_housing.csv', index_col = 0)

cali.head()

plt.figure(figsize = (9, 7))
plt.scatter(cali.latitude, cali.longitude, alpha = 0.2, c = cali.median_house_value, s = cali.population/100)
plt.colorbar()

Looking for Relationships

cali.hist(bins = 40, figsize = (9, 6));

Split the Data

y = cali['median_house_value']
X = cali.drop('median_house_value', axis = 1)
X['total_bedrooms'].fillna(X.total_bedrooms.median(), inplace = True)

X['rooms_per_house'] = X['total_rooms']/X['households']
X['bedrooms_per_room'] = X['total_bedrooms']/X['total_rooms']
X['population_per_household'] = X['population']/X['households']
X_train, X_test, y_train, y_test = train_test_split(X, y)

len(X_train)

train = X_train.join(y_train)
corr_mat = train.corr()

corr_mat['median_house_value'].sort_values(ascending = False)

cali_cols = train[['median_house_value', 'median_income', 'total_rooms', 'housing_median_age']]

scatter_matrix(cali_cols);

lm = LinearRegression()
X = train['median_income'].values.reshape(-1,1)
y = train['median_house_value']
lm.fit(X,y)
predict = lm.predict(X)
mse = mean_squared_error(predict, y)
np.sqrt(mse)

Categorical Data

train.info()

dummies = pd.get_dummies(train['ocean_proximity'])

train = train.join(dummies)

train.columns

train = train.drop('<1H OCEAN', axis = 1)

train.info()

Numerical Features

train = train.drop('ocean_proximity', axis = 1)

train.info()

y = train['median_house_value']
X = train.drop('median_house_value', axis = 1)

Scaling

from sklearn.preprocessing import StandardScaler, PolynomialFeatures

from sklearn.pipeline import Pipeline

pipeline = Pipeline([('std_scaler', StandardScaler()),
                    ('poly_features', PolynomialFeatures(degree = 2))])

housing_prepared = pipeline.fit_transform(X)

Model

lm = LinearRegression()
lm.fit(housing_prepared, y)
lm.score(housing_prepared, y)
lm.predict(housing_prepared[:10])

predictions = lm.predict(housing_prepared)

mse = mean_squared_error(predictions, y)
rmse = np.sqrt(mse)
rmse

Regularized Models, Cross Validation and Grid Search

from sklearn.model_selection import GridSearchCV

lasso = Lasso(max_iter = 100000)
alphas = [0.5, 1, 4, 10, 50, 100]
param_grid = [
    {'alpha': alphas}
]
grid_search = GridSearchCV(lasso, param_grid, cv = 5, scoring = 'neg_mean_squared_error')
grid_search.fit(housing_prepared, y)

grid_search.best_estimator_

lasso = grid_search.best_estimator_

lasso.fit(housing_prepared, y)

predictions = lasso.predict(housing_prepared)
mse = mean_squared_error(predictions, y)
rmse = np.sqrt(mse)
rmse

enet = ElasticNet(max_iter = 100000)
alphas = [0.1, 0.5, 1, 4, 10, 50, 100]
param_grid = [
    {'alpha': alphas}
]
grid_search = GridSearchCV(enet, param_grid, cv = 5, scoring = 'neg_mean_squared_error')
grid_search.fit(housing_prepared, y)

enet = grid_search.best_estimator_
enet.fit(housing_prepared, y)

predictions = enet.predict(housing_prepared)
mse = mean_squared_error(predictions, y)
rmse = np.sqrt(mse)
rmse

Evaluate on Test Set

X_test['total_bedrooms'].fillna(X_test.total_bedrooms.median(), inplace = True)

dummies = pd.get_dummies(X_test['ocean_proximity'])
X_test = X_test.join(dummies)
X_test = X_test.drop('ocean_proximity', axis = 1)
y = y_test

scaler = StandardScaler()
X = scaler.fit_transform(X_test)


enet.fit(X, y)
predictions = enet.predict(X)
mse = mean_squared_error(predictions, y)
rmse = np.sqrt(mse)
rmse

lasso.fit(X, y)
predictions = lasso.predict(X)
mse = mean_squared_error(predictions, y)
rmse = np.sqrt(mse)
rmse

lm.fit(X, y)
predictions = lm.predict(X)
mse = mean_squared_error(predictions, y)
rmse = np.sqrt(mse)
rmse