Allow objects to be passed in as an inputs for network.train

[AkashSamlal]

Allow objects to be passed into an argument for inputs and outputs for network.train

[nicoszerman]

I'm trying to understand your idea. How would the neural network process the input obiect?

[AkashSamlal]

I'm trying to understand your idea. How would the neural network process the input object?

As of currently, we can only pass in a training set as only arrays like this:

However, we should also make it a way for objects to be treated the same, for instance:

[nicoszerman]

Makes sense! I like your idea. We need to think if it's best to add this functionality under .train() or if we should add it under another layer of abstraction.

Nicholas Szerman

On Sun, Jun 30, 2019 at 1:30 PM Akash Samlal notifications@github.com wrote:

I'm trying to understand your idea. How would the neural network process the input object?

As of currently, we can only pass in a training set as only arrays like this: [image: image] https://user-images.githubusercontent.com/43329669/60399931-b34bab80-9b3a-11e9-997b-96480d34abdf.png

However, we should also make it a way for objects to be treated the same, for instance: [image: image] https://user-images.githubusercontent.com/43329669/60399948-0160af00-9b3b-11e9-9ed7-42f78e8a5638.png

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/liquidcarrot/carrot/issues/129?email_source=notifications&email_token=ADLB3ZAV4E7PRD3MMZ4H2ZLP5DUSTA5CNFSM4H4JE2D2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODY4QJ5Q#issuecomment-507053302, or mute the thread https://github.com/notifications/unsubscribe-auth/ADLB3ZB2VFQXY2TJWNKCMCDP5DUSTANCNFSM4H4JE2DQ .

[tracycollins]

This relates to my request to add labels/names to inputs and outputs

[christianechevarria]

Related to #85

Specifically, @tracycollins comment:

Also, an unrelated item that would be nice to have: labels for input and output nodes. I think they are only identified by array index right now? It would help with testing various networks in parallel that have different inputs.

Right now, I add my own wrapper to the network object that provides this feature (among other things).

[christianechevarria]

@tracycollins asking about the wrapper you wrote, is there any feature you incorporated that you would want to see implemented in the library?

[tracycollins]

@christianechevarria The only feature that I found that I really used was the labelling of the inputs/outputs. Thanks!