Run an OLS regression with Pandas Data Frame

问题:

I have a pandas data frame and I would like to able to predict the values of column A from the values in columns B and C. Here is a toy example:

import pandas as pd
df = pd.DataFrame({"A": [10,20,30,40,50], 
                   "B": [20, 30, 10, 40, 50], 
                   "C": [32, 234, 23, 23, 42523]})

Ideally, I would have something like ols(A ~ B + C, data = df) but when I look at the examples from algorithm libraries like scikit-learn it appears to feed the data to the model with a list of rows instead of columns. This would require me to reformat the data into lists inside lists, which seems to defeat the purpose of using pandas in the first place. What is the most pythonic way to run an OLS regression (or any machine learning algorithm more generally) on data in a pandas data frame?

回答1:

I think you can almost do exactly what you thought would be ideal, using the statsmodels package which was one of pandas' optional dependencies before pandas' version 0.20.0 (it was used for a few things in pandas.stats.)

>>> import pandas as pd
>>> import statsmodels.formula.api as sm
>>> df = pd.DataFrame({"A": [10,20,30,40,50], "B": [20, 30, 10, 40, 50], "C": [32, 234, 23, 23, 42523]})
>>> result = sm.ols(formula="A ~ B + C", data=df).fit()
>>> print result.params
Intercept    14.952480
B             0.401182
C             0.000352
dtype: float64
>>> print result.summary()
                            OLS Regression Results                            
==============================================================================
Dep. Variable:                      A   R-squared:                       0.579
Model:                            OLS   Adj. R-squared:                  0.158
Method:                 Least Squares   F-statistic:                     1.375
Date:                Thu, 14 Nov 2013   Prob (F-statistic):              0.421
Time:                        20:04:30   Log-Likelihood:                -18.178
No. Observations:                   5   AIC:                             42.36
Df Residuals:                       2   BIC:                             41.19
Df Model:                           2                                         
==============================================================================
                 coef    std err          t      P>|t|      [95.0% Conf. Int.]
------------------------------------------------------------------------------
Intercept     14.9525     17.764      0.842      0.489       -61.481    91.386
B              0.4012      0.650      0.617      0.600        -2.394     3.197
C              0.0004      0.001      0.650      0.583        -0.002     0.003
==============================================================================
Omnibus:                          nan   Durbin-Watson:                   1.061
Prob(Omnibus):                    nan   Jarque-Bera (JB):                0.498
Skew:                          -0.123   Prob(JB):                        0.780
Kurtosis:                       1.474   Cond. No.                     5.21e+04
==============================================================================

Warnings:
[1] The condition number is large, 5.21e+04. This might indicate that there are
strong multicollinearity or other numerical problems.

回答2:

Note: pandas.stats has been removed with 0.20.0

---

It's possible to do this with pandas.stats.ols:

>>> from pandas.stats.api import ols
>>> df = pd.DataFrame({"A": [10,20,30,40,50], "B": [20, 30, 10, 40, 50], "C": [32, 234, 23, 23, 42523]})
>>> res = ols(y=df['A'], x=df[['B','C']])
>>> res
-------------------------Summary of Regression Analysis-------------------------

Formula: Y ~ <B> + <C> + <intercept>

Number of Observations:         5
Number of Degrees of Freedom:   3

R-squared:         0.5789
Adj R-squared:     0.1577

Rmse:             14.5108

F-stat (2, 2):     1.3746, p-value:     0.4211

Degrees of Freedom: model 2, resid 2

-----------------------Summary of Estimated Coefficients------------------------
      Variable       Coef    Std Err     t-stat    p-value    CI 2.5%   CI 97.5%
--------------------------------------------------------------------------------
             B     0.4012     0.6497       0.62     0.5999    -0.8723     1.6746
             C     0.0004     0.0005       0.65     0.5826    -0.0007     0.0014
     intercept    14.9525    17.7643       0.84     0.4886   -19.8655    49.7705
---------------------------------End of Summary---------------------------------

Note that you need to have statsmodels package installed, it is used internally by the pandas.stats.ols function.

回答3:

I don't know if this is new in sklearn or pandas, but I'm able to pass the data frame directly to sklearn without converting the data frame to a numpy array or any other data types.

from sklearn import linear_model

reg = linear_model.LinearRegression()
reg.fit(df[['B', 'C']], df['A'])

>>> reg.coef_
array([  4.01182386e-01,   3.51587361e-04])

回答4:

This would require me to reformat the data into lists inside lists, which seems to defeat the purpose of using pandas in the first place.

No it doesn't, just convert to a NumPy array:

>>> data = np.asarray(df)

This takes constant time because it just creates a view on your data. Then feed it to scikit-learn:

>>> from sklearn.linear_model import LinearRegression
>>> lr = LinearRegression()
>>> X, y = data[:, 1:], data[:, 0]
>>> lr.fit(X, y)
LinearRegression(copy_X=True, fit_intercept=True, normalize=False)
>>> lr.coef_
array([  4.01182386e-01,   3.51587361e-04])
>>> lr.intercept_
14.952479503953672