Lift some of the limits

[GSPP]

It would be nice to lift some of the limits:

1. Max layers
2. Max neurons per layer
3. Smaller regularization rates
4. Bigger batch sizes
5. More noise

At least lift those limits that are merely constants in the source code and can be changed easily.

[wendli01]

Actually, these restrictions are only enforced in the user interface. You can easily circumvent them by changing the URL.

E.g. in https://playground.tensorflow.org/#activation=tanh&batchSize=10&dataset=gauss&regDataset=reg-plane&learningRate=0.03&regularizationRate=0&noise=2000&networkShape=8,8,8,8,8,8,8,8,8,8&seed=0.78192&showTestData=false&discretize=false&percTrainData=50&x=true&y=true&xTimesY=false&xSquared=false&ySquared=false&cosX=false&sinX=false&cosY=false&sinY=false&collectStats=false&problem=classification&initZero=false&hideText=false

networkShape=8,8,8,8,8,8,8,8,8,8 creates a rather deep network while noise=2000 drastically increases the amount of noise.
The same goes for all of the other parameters you mentioned.

[GSPP]

OK, that's good to know. But I don't see why the limits are so restrictive in the first place. This is a tool for experimentation after all.

Changing them would be a one line change. Are you interested in a PR?

[wendli01]

@GSPP sure, but it is also only a playground.

Anything more involved than what can be configured in the current GUI is probably to resource hungry for the JS implementation anyways and can be experimented with easily enough using the Estimator API.

[JafferWilson]

@wendli01 Sir, will it be possible for you to demonstrate what you say? Regarding the the estimator I am saying. I am looking for a playground example that uses tfjs. Unfortunately, there is no such thing found by me. Can you help?

[GSPP]

What's too resource hungry depends on the machine. The current limits are too much for a phone. They are too small for a normal desktop. Certainly too small for a high-powered developer machine.

I value TF playground very much for the visualization.

I have submitted a PR. I have increased the limits conservatively to address your concerns. I believe this PR gives users what they want so it seems like an improvement to me.

Here is a test link that uses the maximums of the raised limits: http://127.0.0.1:8080/#activation=relu&batchSize=10&dataset=spiral&regDataset=reg-plane&learningRate=0.03&regularizationRate=0&noise=50&networkShape=12,12,12,12,12,12,12,12&seed=0.80450&showTestData=false&discretize=false&percTrainData=50&x=true&y=true&xTimesY=false&xSquared=false&ySquared=false&cosX=false&sinX=false&cosY=false&sinY=false&collectStats=false&problem=classification&initZero=false&hideText=false