arize-phoenix 0.0.1rc0

ML Observability in your notebook

phoenix.ai

CI/CD
Docs
Package
Meta

Phoenix enables you to get to MLOps insights at lightning speed. Phoenix focuses on surfacing areas that require critical attention and offers zero-config observability for model drift, performance, and data quality.

_ NOTE: Phoenix is under active development. APIs may change at any time _

Getting started

pip install phoenix

Troubleshooting

If you are using an Apple M1 machine and encounter the error Could not find a local HDF5 installation, take the following steps:

1. Install HDF5 with arch -arm64 brew install hdf5.
2. Find the path to your HDF5 installation with brew info hdf5.
3. Set an environment variable with export HDF5_DIR=/path/to/your/hdf5/installation.
4. Retry the command you ran when you encountered the error.

If are you using an Apple M1 machine and encounter the error incompatible architecture (have (x86_64), need (arm64e)) during installation (for example, while installing hdbscan), take the following steps:

1. Run softwareupdate --install-rosetta to install Rosetta2 in order to emulate Intel CPUs on your ARM machine.
2. Purge the pip cache with pip cache purge.
3. Retry pip install -e . from the repo base directory.

Unstructured data

Phoenix takes advantage of UMAP and HDBSCAN to highlight segments of your data that are areas of critical drift.

Structured / tabular data

Phoenix surfaces up problematic features of your model with regards to drift, performance, and data quality.