This is somewhat of a theoretical question. I know SO doesn't like code that isn't easily replicated, but please bear with me!
I've got a pandas DataFrame that I want to run a Lasso regression on. To do so, the best way I know of is getting the features into a numpy array:
features = df[list(cols)].values
features = np.nan_to_num(features)
Then I do the sk-learn magic:
lasso_model = LassoCV(cv = 15, copy_X = True, normalize = True, max_iter=10000)
lasso_fit = lasso_model.fit(features, label)
lasso_path = lasso_model.score(features, label)
print lasso_model.coef_
Now my problem is how to efficiently make pandas and numpy work together. This print shows something like:
array([ 1.69066749e-05, -1.56013346e-05, 0.00000000e+00,
-6.77086687e-06, 0.00000000e+00, 3.95920932e-08,
0.00000000e+00, 6.54752484e-06, -0.00000000e+00,
-1.18676617e-05, -7.36411973e-08, 4.72966581e-05,
2.91028626e-06, 1.60674178e-05, 8.83195041e-06,
-8.74769447e-02, 1.39914995e-04, -1.86801467e-05,
3.68593473e-01, 4.16009393e-01, 9.27391598e-07,
-0.00000000e+00, 0.00000000e+00, -4.07446333e-03,
2.33648787e-01, 0.00000000e+00, 2.22660872e-02,
0.00000000e+00, 3.04366897e-02, -0.00000000e+00,
0.00000000e+00, -0.00000000e+00, -0.00000000e+00,
1.85141334e-01, 9.50727274e-02, -4.94268994e-03,
2.22993839e-01, 0.00000000e+00, 1.23715861e-02,
0.00000000e+00, 5.42142052e-02, -1.27412757e-02,
2.98389804e-02, 1.35957828e-02, -0.00000000e+00,
3.64953613e-02, -0.00000000e+00, 1.03289810e-01,])
This does me no good. How do I get what coefficients are for what columns in an efficient manner?
I have found some hack-y ways to do some of it, but I'm thinking there is a much better way that I could do this.
For example, I know I can do the max by:
In [256]: coef.argmax()
Out[256]: 19
In [257]: cols[19]
Out[257]: 'Price'
I think the main thing I'm wondering is how to get a dictionary of column name to coefficient pairs.
Thanks guys!
