Ability to learn by batches of data items

I159 / go_deep

Neural network framework on pure Go

MIT License

1 stars 0 forks source link

Ability to learn by batches of data items #6

Closed I159 closed 6 years ago

I159 commented 6 years ago

Stochastic gradient is overlearning nn very fast. First several tens of data items perform real descent but then loss is kept between 0.33 - 0.67 which leads to a very weak recognition.