automatic testing script and database

[NicoCarpe]

merging testing and master branch

[NicoCarpe]

I'm using 2^x , x=0,1,...,6 for the units and I am making sure that if the units have the form like [2^x,2^x-1,2^x] where the outside terms are always one degree higher.

And I was thinking the next move would be to query the the results such as accuracy and loss against units. However I need to find a way to store units in a way that's not a string so we can properly graph them.

Also I have been using more layers as well and think it might be useful to graph the effects the number of layers have on the accuracy and loss as well.

On Tue., May 28, 2019, 19:52 Kyle Mathewson, notifications@github.com wrote:

@kylemath approved this pull request.

Looks great @NicoCarpe https://github.com/NicoCarpe , now do we visualize the results easily? If I understand you ran a single layer neural network with varying units. This is equivalent (since it is only one layer) to a regression with more and more parameters,

what range of values are you using for the number of units?

Can you plot the test accuracy and test loss as a function of number of units?

@korymath https://github.com/korymath - any other ideas?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/kylemath/DeepEEG/pull/30?email_source=notifications&email_token=AKBLWR7MK3A555LOZRWWGODPXXOUTA5CNFSM4HQH332KYY3PNVWWK3TUL52HS4DFWFIHK3DMKJSXC5LFON2FEZLWNFSXPKTDN5WW2ZLOORPWSZGOBZ57XDY#pullrequestreview-243006351, or mute the thread https://github.com/notifications/unsubscribe-auth/AKBLWRY23QP4WBA3SI3AZQ3PXXOUTANCNFSM4HQH332A .

[kylemath]

sounds great, yes manipulating layers is a good idea too,

there is no need to have fewer units in the inner layers, so feel free to varry that as well,

On Tue, May 28, 2019 at 10:00 PM Nicolas Carpenter notifications@github.com wrote:

I'm using 2^x , x=0,1,...,6 for the units and I am making sure that if the units have the form like [2^x,2^x-1,2^x] where the outside terms are always one degree higher.

And I was thinking the next move would be to query the the results such as accuracy and loss against units. However I need to find a way to store units in a way that's not a string so we can properly graph them.

Also I have been using more layers as well and think it might be useful to graph the effects the number of layers have on the accuracy and loss as well.

On Tue., May 28, 2019, 19:52 Kyle Mathewson, notifications@github.com wrote:

@kylemath approved this pull request.

Looks great @NicoCarpe https://github.com/NicoCarpe , now do we visualize the results easily? If I understand you ran a single layer neural network with varying units. This is equivalent (since it is only one layer) to a regression with more and more parameters,

what range of values are you using for the number of units?

Can you plot the test accuracy and test loss as a function of number of units?

@korymath https://github.com/korymath - any other ideas?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub < https://github.com/kylemath/DeepEEG/pull/30?email_source=notifications&email_token=AKBLWR7MK3A555LOZRWWGODPXXOUTA5CNFSM4HQH332KYY3PNVWWK3TUL52HS4DFWFIHK3DMKJSXC5LFON2FEZLWNFSXPKTDN5WW2ZLOORPWSZGOBZ57XDY#pullrequestreview-243006351 , or mute the thread < https://github.com/notifications/unsubscribe-auth/AKBLWRY23QP4WBA3SI3AZQ3PXXOUTANCNFSM4HQH332A

.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/kylemath/DeepEEG/pull/30?email_source=notifications&email_token=AA36GFKHBELEDUAAC6KN5J3PXX5W7A5CNFSM4HQH332KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWOC5LI#issuecomment-496774829, or mute the thread https://github.com/notifications/unsubscribe-auth/AA36GFMO4OMJBSW2NY64TZ3PXX5W7ANCNFSM4HQH332A .

[kylemath]

Hey @NicoCarpe just saw this old PR, how are you doing lately?

[NicoCarpe]

Hi, I've been doing alright considering the situation (not the biggest fan of the online class format). How about yourself?

Also, after you introduced me to this project of yours I became quite interested in computer science applications to neuroscience, but I felt a bit under-equipped to help out properly at the time. In the past few semesters I have taken classes that I felt filled in those gaps of knowledge but I still haven't really gotten my foot in the door in terms of research.