redshells

redshells copied to clipboard

to follow Python 3.8+

hirosassa

`classification_report` on `fit data` is almost worthless. Maybe, this log will missleading. line: https://github.com/m3dev/redshells/blob/master/redshells/train/utils/utils.py#L16 For example. ``` [2020/03/03 23:22:07][redshells.train.utils.utils][INFO](utils.py:21) precision recall f1-score support 0 1.00 1.00 1.00 3060 1 1.00...

vaaaaanquish

warning now comes out. > W1022 09:22:52.705346 4633392576 deprecation_wrapper.py:119] From python3.6/site-packages/redshells-0.1.7-py3.6.egg/redshells/model/early_stopping.py:19: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.

vaaaaanquish

`matrix[user_indices, item_indices]` in graph_convolutional_matrix_completion.py L.404 returns error in scipy==1.3.1, since indexing with numpy.ndarray is not possible. Possible Solution - `matrix[len(user_indices), len(item_indices)]` or - specific scipy version as `scipy==1.2.1` at redshells/setup.py

mski-iksm

When flag `with_user_embedding` is False, using default user index would be better rather than using index zero.

Teppei-Kanayama