airbnb_polyreg_R.py

import matplotlib.pyplot as plt
%matplotlib inline
from sklearn.metrics import r2_score

# importing R functions
#!pip install rpy2
import rpy2.robjects as robjects
r_predict = robjects.r["predict"]
r_lm = robjects.r["lm"]

df_r2_scores = pd.DataFrame({'R2':[]})

# function that fits a linear model on training dataset, plots the regression line and report the R2 on test
# The 'order' parameter can be specified to add higher order polynomial terms
def fit_polynomial(order, df_train_summary=df_train_summary, df_test_summary=df_test_summary, title='model'):
    # populating vectors R understands
    r_date_train = robjects.FloatVector(df_train_summary.date) 
    r_price_train = robjects.FloatVector(df_train_summary.price)
    # populating dataframe R understands
    bookings_train_r = robjects.DataFrame({'date': r_date_train, 'price': r_price_train})
    # populating formula R understands
    simple_formula = robjects.Formula("price ~ poly(date, {}, raw=TRUE)".format(order))
    # running models in R
    bookings_lm = r_lm(formula=simple_formula, data=bookings_train_r)
    # getting prediction line in R
    predict_df = robjects.DataFrame({'date': robjects.FloatVector(np.linspace(0, 364, 100))})
    predictions = r_predict(bookings_lm, predict_df)
    # getting results back to Python and plotting prediction line for train data
    fig, ax = plt.subplots(1,2,figsize=(15,5))
    df_train_summary.plot.scatter(x='date',y='price',c='Red',title="{} - Bookings train data".format(title), ax=ax[0])
    ax[0].set_xlabel("date")
    ax[0].set_ylabel("price");
    ax[0].plot(predict_df.rx2("date"),predictions);
    # getting test predictions in R
    predict_test_df = robjects.DataFrame({'date': robjects.FloatVector(df_test_summary.date)})
    predictions_test = r_predict(bookings_lm, predict_test_df)
    # getting results back to Python and plotting prediction line for test data
    df_test_summary.plot.scatter(x='date',y='price',c='Green',title="{} - Bookings test data".format(title), ax=ax[1])
    ax[1].set_xlabel("date")
    ax[1].set_ylabel("price");
    ax[1].plot(predict_test_df.rx2("date"),predictions_test);
    # computing R2 score
    r2 = r2_score(df_test_summary.price, predictions_test)
    print(title)    
    print("R2 on test: {}".format(r2))
    df_r2_scores.loc[title] = r2
    
fit_polynomial(5, title='5-order Polynomial')
fit_polynomial(25, title='25-order Polynomial')
fit_polynomial(50, title='50-order Polynomial')