raivnlab / fluid Goto Github PK

Please see this SO issue for an overview of the issue.

Could you please update with a working pip environment?

Hi,

Thanks for open sourcing your code! I really enjoyed going through your paper. I have a few questions I was hoping you could answer.

1. In the paper you mention that
   For all experiments that require training, ..., we train each model for 4 epochs every 5,000 samples that are received.
   If I understand correctly, this means methods like Standard Training and Finetuning will stop and train 90k / 5k = 18 times during the sequential phase ? And other methods like Prototype Finetuning are updated on-the-fly without any offline training ?
2. Could you give a little more info on how the samples in the stream are ordered ? Are the samples ordered in any way (e.g. all samples from class 1 are showed before moving on two class 2) or is it possible (although super unlikely but still) that the first 1K samples contain all 1K classes ?
3. Finally, you mention in the readme that there exists a file called overview.md which lays out the code structure ? Could you link me to the file (as I could not find it) ?

Again, thanks for sharing your code!