gbuesing / neural-net-ruby Goto Github PK

I think a nice addition will be to pass in an option the if mse: %0.01 or lower\higher didn't change between iterations in the learning step.

Does that makes sense ? or do you think this is better handled outside the class ?

Dear Geoff,

we've recently added neural-net-ruby to our RubyML list: https://github.com/arbox/machine-learning-with-ruby

In order to connect all Ruby & ML related projects we introduced the rubyml topic on GitHub.

Please cosider adding the rubyml topic to this repository as well so that the Ruby community can discover neural-net-ruby faster 😃

Thank you in advance!

Hi,
As this repo has no license I'm not sure what is allowed here, I want to port this lib to Crystal (https://crystal-lang.org/) Is that ok with you ?

Is there a way to make this lib predict text classifications ?
Very much like byes-classifiers but using a pre-ran learning phase with 2 data sets ?

So, In my example I let the nn instance to learn on a set of data, after the learning stage I save the ANN into a file.
Later when I load it and run it again it seems to "forget" what it learns and the error reach back to mse 44% from 3.09%

After training:

Trained classification success: 253,failure: 13(classification error: 4.887218045112782%, mse: 3.09%)

After loading and re-running:

Untrained classification success: 141,failure: 125(classification error: 46.99248120300752%, mse: 42.25%)

*Ignore the "Untrained" part ;)

Am I missing something ?