alibi-detect
alibi-detect copied to clipboard
Algorithms for outlier, adversarial and drift detection
Hello, I have a question regarding the consistency of the MMD Tensorflow detector and the MMD PyTorch detector in the [`cd_text_imdb.ipynb` Jupyter Notebook](https://github.com/SeldonIO/alibi-detect/blob/cb512067bc8206816eeddff37f004ef492ccd72c/doc/source/examples/cd_text_imdb.ipynb). In my opinion, the preprocessing function passed...
The `requests.get()` calls in the fetch methods such as `fetch_nab` and `fetch_cifar10c` currently don't have any error handling, potentially resulting in unclear errors when the requested url is down. For...
The current implementation of the SR outlier detector does not support non-uniformed sampled signals. This is due to FFT. A potential solution might be to investigate NFFT described in [here](https://www-user.tu-chemnitz.de/~potts/paper/nfft3.pdf)...
In a number of places (such as `GaussianRBF`, and for detectors with backends), kwargs default to `None`, and then are set to backend-specific objects internally. We should survey all these...
We should explore adding a base drift detector class. This could help with a number of issues: 1. The `preprocess` methods are heavily duplicated across drift detectors (in `cd.base`): ```python...
Detectors with a backend, such as `MMDDrift` inherit their backend-specific subclass names in their `self.meta` dict i.e, `MMDDriftTF`. The top-level detector name i.e. `MMDDrift` should instead be stored. See https://github.com/SeldonIO/alibi-detect/pull/497#discussion_r880276630.
The drift detection submodule was named `cd` for "concept drift". We should rename it to `dd` for drift detection. There are also a number of items scattered around that could...
https://github.com/SeldonIO/alibi-detect/blob/b4a62f3c64af21ba5f6fd35eb57b2d3af0c6665a/alibi_detect/od/prophet.py#L123 Any kind of kwards for stan init can be added here to make warm start
I have trained a VAR model and i would like to convert it to onnx to deploy on an edge device. Is it possible to use tfonnx conversion for this...
After the change to `sphinx-autoapi` (https://github.com/SeldonIO/alibi-detect/pull/482), there are some new objects showing up in the api docs which are undesired, for example the `logger`: ![image](https://user-images.githubusercontent.com/32061685/163418971-33db606f-720a-40f5-b873-bfb2c565e1e7.png) This is a symptom of...