Add formula to stat_smooth

[has2k1]

• BUG: Fix accepting of eval environment for formula statsmodels/statsmodels#6152 Has been merged
• BUG: Fix accepting of eval environment for formula statsmodels/statsmodels#6152 Has been released
• Minimum version of statsmodels has been upgrade to include BUG: Fix accepting of eval environment for formula statsmodels/statsmodels#6152
• Tests pass

closes #311

[has2k1]

@MartinGell

With this you would be able to use fomulae to define models for linear regressions. A good (orthogonal) poly method stagnated is in the patsy PR queue, but for non analytical work a basic one would do.

import numpy as np
import pandas as pd
from plotine import *

n = 100
np.random.seed(123)
mu = 0
sigma = 15
noise = np.random.randn(n)*sigma + mu

def poly(x, degree=1):
    """
    Fit Polynomial

    These are non orthogonal factors, but it may not matter if
    we only need this for predictions (without interpreting the
    coefficients) or visualisation.
    """
    d = {}
    for i in range(degree+1):
        if i == 1:
            d['x'] = x
        else:
            d[f'x**{i}'] = np.power(x, i)
    return pd.DataFrame(d)

x = np.linspace(-np.pi, np.pi, n)
df = pd.DataFrame({
    'x': x,
    'y': (x + x**2 + 7*x**3) + noise
})

(ggplot(df, aes('x', 'y'))
 + geom_point()
 + stat_smooth(method='lm', formula='y ~ poly(x, degree=3)', fill='red', se=True)
)

[krassowski]

@has2k1 I see that your PR to statsmodels (6152) has been merged and statsmodels 0.11 released. Is there anything else holding this one out? Could we help in any way?

[has2k1]

@krassowski, with the release of statsmodels v0.11 then this PR will get merged.