sticker

sticker copied to clipboard

QoL improvement: Set best epoch in config after train

3

Maybe for sticker2, if it's too invasive for sticker?

sebpuetz

Writing the labels to yml or toml can still be done via serde but allows for manual inspection.

twuebi

We could bucket sequences in the dev set by their length to speed up dev passes

twuebi

danieldk

Make the exponential decay lr schedule available

6

Right now, the exponential decay lr schedule is not available for `sticker train`and `sticker pretrain`. Once #145 is merged, it would make sense to have it available for both subcommands....

twuebi

The `conllx` dependency is now used throughout `sticker`. Abstract over layers and features using traits, so that the `sticker` create becomes oblivious to the underlying format. `sticker-utils` could then provide...

danieldk

Support of custom features

1

Hello, Thank you for this awesome library. I´m very impressed by the quality. My task is to extract/segment information from a semi structured text. But not only the text is...

bratao

Epoch summaries on tensorboard

3

Add epoch-averaged (train and dev) values of the summarized metrics to tensorboard. After a few epochs it's hard to tell anything from the per-batch graphs.

sebpuetz

Add gradient norm to train output

4

Printing the gradient norm alongside the training output can be helpful when tuning learning rates and for finding the gradient clipping threshold.

twuebi

Once an epoch is done, we start a new line. The batch accuracy/loss that is displayed is then that of the last batch of the epoch. These numbers are confusing...

danieldk