Scott Sievert

Sorry for the delayed response. I'll have more time to respond a week from now. > Hyperband doesn't appear to be handling the validation data correctly Could the issue be...

> I can't tell if it's handling my training data correctly either From what I can tell, you're handling the test data correctly: it appears you're only running the test...

I have one more basic question: * Why does model convergence depend on "pad[ding] at the batch level) vs. padding to the longest example in the dataset"? Is "convergence" in...

Is an average coefficient difference of 0.008 large? That's what `np.abs(mod_sklearn.coef_ - mod_dask.coef_).mean()` would produce. Yes, the sum is large (8) but the number of examples is much larger (1000)....

Hm... I've modified the script a bit below, and get different results when I train Dask-ML's `LogisticRegression` on NumPy arrays and a Dask array of the same data: ``` shell...

Thanks for that output @kchare. I think I've found the issue: 1. Dask-GLM's `utils.add_intercept` adds coefficients representing the intercept to be the first or last column if Dask or NumPy...

I'd love to see a PR for that! I think it'd be a valuable PR, mostly because I think Dask-ML should have a test showing it converges to Scikit-learn's solution...

> Better scheduling (any examples here?) I fit and score serially in https://github.com/dask/dask-examples/pull/15: ``` python inc = Incremental(...) data = [] for i in range(40): inc.fit(X, y) data += [{'score':...

This PR is motivated by a question from @amueller after [my SciPy19 talk]. Why run this bracket if it's not the best performing? In simulations, `bracket=0` is the best bracket...

> what if you only ran bracket 4 [the most aggressive bracket] in all cases? I'll take your question as "how can the search be made more aggressive?" The aggressiveness...