import pandas as pd
import numpy as np

%load_ext rmagic

%%R -o data
set.seed(3433); par(mfrow=c(1,2)); 
data <- rnorm(100,mean=seq(0,3,length=100),sd=seq(0.1,3,length=100))

df = pd.DataFrame(zip(data, np.linspace(0, 3, num=100)), columns=['data', 'x'])

from statsmodels.formula.api import ols
from statsmodels.stats.sandwich_covariance import cov_hc3

lm1 = ols('data ~ x', df).fit()

cov_hc3(lm1)

lm1.normalized_cov_params

%%R -o x,y
set.seed(343)
x <- seq(0,3,length=100); y <- rcauchy(100)

df = pd.DataFrame(zip(x, y), columns=['x', 'y'])

from statsmodels.formula.api import rlm

lm1 = ols('y ~ x', df).fit()
rlm1 = rlm('y ~ x', df).fit()

lm1.params

rlm1.params

f, (ax1, ax2) = subplots(ncols=2)

ax1.plot(df['x'], df['y'], 'o', color='grey')
ax1.plot(df['x'], lm1.fittedvalues, linewidth=3)
ax1.plot(df['x'], rlm1.fittedvalues, 'g', linewidth=3)
ax1.set_xlabel('x')
ax1.set_ylabel('y')

ax2.plot(df['x'], df['y'], 'o', color='grey')
ax2.plot(df['x'], lm1.fittedvalues, linewidth=3)
ax2.plot(df['x'], rlm1.fittedvalues, 'g', linewidth=3)
ax2.set_xlabel('x')
ax2.set_ylabel('y')
ax2.set_ylim([-5, 5])
ax2.set_title('Zoomed in')

f.set_size_inches(9, 3)
f.tight_layout();

movies = pd.read_csv('../data/movies.txt', sep='\t')
movies.columns = ['X', 'score', 'rating', 'genre', 'box_office', 'running_time']
movies.head()

# no stepwise regression
lm1 = ols('score ~ box_office + running_time', movies).fit()
lm1.summary()

# no regsubsets
# no bic.glm