Ps in ConvLearner.pretrained

What does ps mean? In dogsbreed, the ps is set to 0.2 at some point. Thanks!

1 Like

It will be explained in later lessons (depending on where you are in the video).
In short, it refers to the dropout parameter. This is a parameter that helps combat overfitting by telling the learner that for certain layers, during training, it should throw away 20% of the activations.
I recommend watching the lesson (3 or 4?) where Jeremy demonstrates how NNs work in MS Excel. He also shows there how dropout works.

Thanks for the reply. I already know what dropout is, just that the name of the parameter, ps, isn’t immediately obvious to me of what it refers to.

1 Like