the-algorithm-ml
the-algorithm-ml copied to clipboard
Source code for Twitter's Recommendation Algorithm
Figured this should have a tracking issue. See https://github.com/twitter/the-algorithm-ml/blob/78c3235eee5b4e111ccacb7d48e80eca019e480c/common/run_training.py#L87 & https://github.com/pytorch/pytorch/issues/79388 Spotted via #50 ( thanks, @guibaq )
Related to #48
TODO: - [X] Configure `mypy`. - [X] Configure `isort` to make imports more readable. - [X] Add `mypy` on the `common` package. - [x] Add `mypy` as a `pre-commit` hook...
$ ./projects/home/recap/scripts/create_random_data.sh This will create some random data (in $HOME/tmp/recap_local_random_data). $ ./projects/home/recap/scripts/run_local.sh replace with $ ./projects/home/recap/script/create_random_data.sh This will create some random data (in $HOME/tmp/recap_local_random_data). $ ./projects/home/recap/script/run_local.sh
Minuscule nitpicks. 1. Bad initialization gain value for ReLU nets. The initial weights for this feedforward block https://github.com/twitter/the-algorithm-ml/blob/78c3235eee5b4e111ccacb7d48e80eca019e480c/projects/home/recap/model/mask_net.py#L39 are set using https://github.com/twitter/the-algorithm-ml/blob/78c3235eee5b4e111ccacb7d48e80eca019e480c/projects/home/recap/model/mask_net.py#L10 This will initialize the block with uniform weights...
Fix comment formatting in losses.py
Some light fixes to the heavy ranker README to make it more readable.
Edit typo.
I think in each case the second definition overwrites the initial definition??? at least that is what I would expect. projects/home/recap/models/config.py BatchNormConfig is defined twice DropoutConfig is defined twice MlpConfig...