Tidied up hidden_layers

[JanPokorny]

Hi, I have tidied up the function hidden_layers:

• fixed the inconsistency with required parameters
• removed the ugly, slow, and confusing-stacktrace-producing misuse of try/catch
• small PEP8 fixes

The test coverage should stay the same, as the functionality didn't change.

[pep8speaks]

Hello @JanPokorny! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! :beers:

Comment last updated at 2019-07-17 04:07:04 UTC

[coveralls]

Pull Request Test Coverage Report for Build 497

• 7 of 8 (87.5%) changed or added relevant lines in 1 file are covered.
• 6 unchanged lines in 4 files lost coverage.
• Overall coverage increased (+1.3%) to 84.727%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
talos/model/hidden_layers.py	7	8	87.5%
<!--	Total:	7	8	87.5%	-->

Files with Coverage Reduction	New Missed Lines	%
talos/init.py	1	95.24%
talos/model/early_stopper.py	1	88.89%
talos/commands/deploy.py	2	92.45%
talos/scan/scan_addon.py	2	93.1%
<!--	Total:	6		-->

Totals
Change from base Build 500:	1.3%
Covered Lines:	1054
Relevant Lines:	1244

---

💛 - Coveralls

[mikkokotila]

Thanks but I think there is an issue.

For the common case where there is no kernel_initializer in params dictionary, the proposed:

kernel_regularizer=params.get('kernel_regularizer')

returns nothing. Whereas, the original:

    try:    
        kernel_initializer = params['kernel_initializer']   
    except KeyError:    
        kernel_initializer = 'glorot_uniform'

...returns glorot_uniform ... which is exactly what we want because it's the Keras default setting for the argument.

If there is a better way to do this (ideally have Talos automatically use Keras default setting if not set in the params dictionary), then that would be great.

[JanPokorny]

@mikkokotila Hello, you have confused kernel_initializer and kernel_regularizer! Take a second look, my code does precisely what the old did. Note that .get(...) returns the value of its second argument when the value is not present in the dictionary (or None if you don't specify it).

[mikkokotila]

Ah, got it, that's right I was confused indeed 👍

Ok I understand now, in that case this is great. Can you make the PR to daily-dev as that's the place for PR merges. It will make its way from there to dev and eventually master (and finally at some point to pypi i.e. production).

[JanPokorny]

@mikkokotila I was following CONTRIBUTING.md, you should update it then!

[JanPokorny]

@mikkokotila Also unsure why Travis is broken, it was broken even before I changed the destination branch. Looks like some problem with pickling. I think that this PR didn't cause it as it essentially doesn't change any behavior. Can you check that please?

[mikkokotila]

Sorry for taking ages with this. The big push to move from 0.5 to 0.6.3 took a couple of months in total, and this came a couple of weeks into that. Thanks again :)