FIX: OLS.predict() - cast input to DataFrame from Series before performing operations

[rockg]

Closes #5233. Simply cast to a Series to a DataFrame before dropping NAs or filling.

[jreback]

can u add some tests that demonstrate the issue?

[jtratner]

Frankly, I think the better fix (separate from this PR) would be to have Series' dropna match signature of DF and Panel, since both 'any' and 'all' or thresh apply. Just means that you don't have to care about what kind of NDFrame you have.

Please update your commit message to emphasize that it's only in OLS-Predict. I read the title and was quite confused.

[rockg]

@jreback @jtratner Here is the issue demonstrated and I put this as a test in the last commit that now passes with the fix. I agree with @jtratner that series and dataframes should have the same drop arguments.

y = tm.makeTimeSeries()
x = tm.makeTimeSeries()
model = ols(y=y, x=x)
model.predict(x=x)
Traceback (most recent call last):
  File "<console>", line 1, in <module>
  File ...ols.py, line 446, in predict
    x = x.dropna(how='any')
TypeError: dropna() got an unexpected keyword argument 'how'

[jtratner]

looks fine to me.

[jreback]

closing in favor of #5251 (credited for your tests!)