%pylab inline --no-import
import pylab as pl
import seaborn as sns
import numpy as np 
from scipy import stats
import pandas as pd
from IPython.display import HTML, display

from sklearn.preprocessing import label_binarize, LabelEncoder

sns.set(context = 'notebook', style = 'whitegrid')

x = np.linspace(0, 14, 100)
y1 = np.sin(x + .5)
y2 = np.sin(x + 4*.5) * 3
#c1, c2 = sns.color_palette("deep", n_colors = 2, )
pl.plot(x, y1)
pl.fill_between(x, y1 -.5, y1+.5, alpha = .2)
pl.plot(x, y2)
pl.fill_between(x, y2 -.5, y2+.5, alpha = .2)

weather = pd.read_csv('../data/weather.csv', parse_dates=['Date'])
weather.columns

sns.boxplot(weather.loc[:, ['MinTemp', 'MaxTemp', 'Humidity9am', 'Humidity3pm']])

sns.boxplot(weather.loc[:, 'MinTemp'], groupby=weather.RainTomorrow)
pl.title('MinTemp ~ RainTomorrow')

sns.regplot(x = weather.Pressure9am, y = weather.Temp9am,
            xlabel='Pressure9am', ylabel = 'Temp9am',)

## A more data frame friendly way
sns.regplot(x = 'Pressure3pm', y = 'Temp3pm', 
            data = weather, corr_func=stats.kendalltau,)

sns.lmplot(x = 'Pressure3pm', y = 'Temp3pm', color = 'RainToday', order = 2,
            data = weather, )

sns.lmplot(x = 'Pressure3pm', y = 'Temp3pm', col = 'RainToday', fit_reg=False, 
            data = weather, )

sns.lmplot(x = 'Pressure3pm', y = 'Temp3pm', col = 'RainToday', color = 'RainTomorrow', 
            data = weather, )

sns.lmplot(x = 'Pressure3pm', y = 'Temp3pm', col = 'RainToday', row = 'RainTomorrow', 
            data = weather, )

df = weather.copy()
df['RainTodayEncoded'] = LabelEncoder().fit_transform(df.RainToday)
sns.lmplot(x = 'RainTodayEncoded', y = 'Temp3pm', color = 'RainTomorrow',
            data = df, )

df = weather.copy()
df['RainTomorrowBinary'] = label_binarize(weather.RainTomorrow, classes=['Yes', 'No']).ravel()
sns.lmplot(x = 'Pressure3pm', y = 'RainTomorrowBinary', 
           logistic=True, 
            data = df, )

## partial regression of Temp9am, Temp3pm, conditioning on MaxTemp
sns.lmplot('Temp9am', 'Temp3pm', weather)
sns.lmplot('Temp9am', 'Temp3pm', weather, x_partial='MaxTemp')

pl.figure(figsize=(25, 25))
sns.corrplot(weather)

sns.coefplot('Temp3pm ~ RainToday + Pressure3pm + RainToday * Pressure3pm + MinTemp + MaxTemp + MinTemp * MaxTemp', 
             data = weather, intercept=True)

print stats.skew(weather.loc[:, ['MinTemp', 'MaxTemp']])

sns.kdeplot(weather.loc[:, ['MinTemp']], label = 'MinTemp')
sns.kdeplot(weather.loc[:, ['MaxTemp']], label = 'MaxTemp')
pl.title('kde')
pl.legend(loc = 'best')

## estimate lambda for Box-Cox transform
for lamb in np.linspace(-1, 1, 50):
    print lamb, stats.skew(map(lambda x: (x**lamb - 1.)/lamb, np.array(weather.loc[:, ['MaxTemp']])))

sns.kdeplot(map(lambda x: (x**.3 - 1.)/.3, np.array(weather.loc[:, ['MaxTemp']])), label = '$log(MaxTemp)$')
print stats.skew(map(lambda x: (x**.3 - 1.)/.3, np.array(weather.loc[:, ['MaxTemp']])))

n_pts = 29
x = np.linspace(0, 14, n_pts)
true_data = np.sin(x)
np.random.seed(9221999)
n_subjs = 20
subj_noise = np.random.rand(20) * 2
subj_data = np.array([true_data +  # real signal
                      np.random.randn() +    # subject specific offset from real signal
                      np.random.randn(n_pts) * (subj_noise[s])  # sample specific error with subject specific variance
                      for s in range(n_subjs)])
sns.tsplot(x, subj_data, err_style="ci_bars")

df = weather.copy()
df.sort('Date')
sns.tsplot(df.Date, df.Cloud3pm, interpolate=False, )