NEAT class documention about carrot

[dan-ryan]

Looks fine on Chrome now. Also tested Edge.
If I was being picky... The changes had a number of CSS lines commented out without comments :)

[christianechevarria]

Hey guys, I went ahead and exposed a basic NEAT entry in the docs with 2306d3a

Extending network.evolve sounds great and it's something I want to work on in the coming days

[dan-ryan]

Looks like the layout is broken on Chrome.

[christianechevarria]

If I was being picky... The changes had a number of CSS lines commented out with

Great catch! Went ahead and cleaned it up with 91fe43a

Love the "pickiness" definitely keep it coming lol

[luiscarbonell]

@dan-ryan Thank you for pointing it out! 👍 @wagenaartje had built the evolutionary greatness of Neat into the network.evolve function - I just went ahead and documented it here.

That said, I 100% agree that we should document the Neat class somehow. @christianechevarria and I had talked about maybe just dealing with "1" network (i.e. Network.evolve) or creating a more descriptive name like Population, Species, etc. to describe the function of the Neat class in Neataptic and using other evolutionary algorithms or populations/multi-network methods in that class.

I will leave this issue open for now, until we converge on some good concept/architecture for this one.

Feedback, opinions, preferences, are 💯 welcome!

[dan-ryan]

I was going to write this in another issue, but since you asked for feedback here...

I have found the Neat class is much more complicated than it needs to be.

My ideal api would be:

• You extend network.evolve.
• You pass it a score that for the network to hit to finish.
• You pass in the scoring function.
• Select the population class. So you can extend and store data if needed.
• The population has ids.
• Multi-threading support.
• Able to give it your data and select what percentage is the training, validation and testing data.
• A setting that auto adjusts the mutation rate based on the validation results.
• Regularization support.

[luiscarbonell]

@dan-ryan I would 100% agree with this.

I was going to write this in another issue, but since you asked for feedback here...

I have found the Neat class is much more complicated than it needs to be.

My ideal api would be:

You extend network.evolve.
You pass it a score that for the network to hit to finish.
You pass in the scoring function.
Select the population class. So you can extend and store data if needed.
The population has ids.
Multi-threading support.
Able to give it your data and select what percentage is the training, validation and testing data.
A setting that auto adjusts the mutation rate based on the validation results.
Regularization support.

[christianechevarria]

@dan-ryan Just pushed up some changes with 371790b

Did some basic tests on my end for different viewports, but want to check with you guys before closing this issue and #32

Let me know what you guys think!