Closed umairjavaid closed 4 years ago
We do not have any code for it. In fact, the random hyperparameter search algorithm is too simple to write a code ;)
You can generate configs (arguments) for train.py
with your favourite random number generator according to the search space indicated in Table 5 (appendix). For example, we have used the random number generators in Google Sheets to generate the configs. You can also use Numpy random number generators.
Afterwards, you choose the configs with the best model performance as your final model, which again does not really require much coding.
As mentioned in the paper, where is the code for hyperparameter search?