alibi-detect
alibi-detect copied to clipboard
Algorithms for outlier, adversarial and drift detection
Looks like they've been on 1.x for a while: https://github.com/facebook/prophet/releases This is in conjunction with `holidays` update: #623.
Settle on a consistent way to parameterise tests: direct parameterisation (preferable) vs. the pattern below (mainly used throughout the library so far). ```python tests = list(product(n_features, n_instances, kernel_a, kernel_b, eps))...
There is an error in [this table](https://docs.seldon.io/projects/alibi-detect/en/stable/overview/saving.html#supported-detectors): ![image](https://user-images.githubusercontent.com/32061685/188863616-6050f998-f6d1-46b6-9f79-f6756bd6bc85.png) The `CVMDrift` and `FETDrift` detectors do not have legacy save/load support.
I have created a KSDrift based detector (referred https://docs.seldon.io/projects/alibi-detect/en/stable/examples/cd_text_imdb.html) While saving the detector using save_detector function, I am getting the below error: --------------------------------------------------------------------------- NotEncodableError Traceback (most recent call last) ~\Anaconda3\lib\site-packages\alibi_detect\saving\saving.py...
#619 explicitly sets our CI to use `Python 3.10.6` until [this issue](https://github.com/python/mypy/issues/13627) is fixed. We should monitor the upcoming `mypy 0.980` and `0.990` releases and revert to setting `3.10` in...
Updates the requirements on [tensorflow](https://github.com/tensorflow/tensorflow) to permit the latest version. Release notes Sourced from tensorflow's releases. TensorFlow 2.10.0 Release 2.10.0 Breaking Changes Causal attention in keras.layers.Attention and keras.layers.AdditiveAttention is now...
Updates the requirements on [tensorflow-probability](https://github.com/tensorflow/probability) to permit the latest version. Release notes Sourced from tensorflow-probability's releases. TensorFlow Probability 0.18.0 Release notes This is the 0.18.0 release of TensorFlow Probability. It...
The reason is that the optimal batch size for training can be wildly different than that for prediction (where we just care about being as fast as possible within our...
The static `trainer` method for the PyTorch and KeOps backends of LearnedKernelDrift are the same and could be shared.
Allow to change the `num_workers` argument for the torch dataloaders where relevant.