search

netrack / keras-metrics

Metrics for Keras. DEPRECATED since Keras 2.3.0
MIT License
165 stars 23 forks source link

Upgrading keras gives decimal values for true_positive, false_positive, true_negative, false_negative #50

Open amolgharat opened 4 years ago

amolgharat commented 4 years ago

After upgrading Keras from version 2.2.5 to 2.3 I get decimal values for true_positive, false_positive, true_negative, false_negative. It works fine with version 2.2.5. Also, Keras version 2.3 has added new metrics like true_positive, false_positive, precision, recall etc. However, these metrics values don't match between Keras and keras-metrics.

ybubnov commented 4 years ago

@amolgharat, true, seems builds (and tests) are failing for Keras 2.3.0 and tensorflow 2.0.0.

ybubnov commented 4 years ago

I've added in #49 a deprecation warning for this library that support will be dropped since Keras 2.3.0 (as they already have a replacement for all metrics).

monitarb commented 4 years ago

Please son't stop the good job you were doing. Keras Metrics in version 2.3.1 don´t work properly with to_categorical classes (i.e One hot encoded [0,1] instead of just 1). False Positives and False Negatives give the same result (don't know why), so accuracy and recall give the same result as well. I downgraded my keras version to 2.2.5 just so I could use your library that gives correct results. Please, keep up the good work!