I all the time stay up for sharing these updates with you! Hope you’ll discover one thing right here that may improve your workflow. Right here’s what we’ve launched within the final quarter.
New 🎉
1. MosaicML Composer integration
New integration alert! With the Neptune-Composer integration, you possibly can mechanically log your Composer coaching metadata to Neptune.
See the instance venture right here.
2. Assist for Seaborn figures
We’re increasing the record of supported visualizations. Now, you possibly can log and show Seaborn figures as PNG photographs.
3. Documentation updates
We documented the Neptune Question Language, which can be utilized to filter runs or fashions queried from a venture. This characteristic is experimental. So, we’re blissful to listen to your suggestions by way of GitHub!
Oh, and by the way in which, we’d actually admire it for those who gave us a star and helped develop our repo.
We additionally printed a couple of new guides and tutorials.
Monitoring fashions end-to-end: This one exhibits how one can leverage Neptune to trace a mannequin throughout all phases of its lifecycle (together with logging metadata, evaluating runs, and monitoring the mannequin in manufacturing).
Utilizing a number of integrations without delay: Right here, we’re addressing the query of whether or not you possibly can profit from a number of Neptune integrations on the similar time. The reply is sure. Test the information for an instance.
See different documentation information right here.
Enhancements 🤓
Python shopper library
We took a step towards matching the online app’s querying capabilities: When fetching runs with fetch_runs_table(), you’ve 4 new parameters: sort_by, ascending, restrict, and progress_bar. Additionally applies to model-fetching strategies: fetch_models_table() and fetch_model_versions_table(). (neptune 1.9.0)
Progress bars at the moment are enabled by default when fetching metadata through API. It’s also possible to outline and use your personal bar. (neptune 1.9.0)
All messages that Neptune prints to the console are prefixed with [neptune]. (neptune 1.9.0)
We launched querying performance to desk fetching strategies. See the Documentation updates part above. (neptune 1.10.0)
We enabled the auto-scaling of pixel values when logging figures with the as_image() methodology. (neptune 1.10.0)
Integrations
We launched the log_model_summary parameter to NeptuneCallback() ( neptune-tensorflow-keras 2.2.1)
We added assist for logging venture necessities with the dependencies parameter. (kedro-neptune 0.4.0)
We up to date the combination to make use of the get_state() methodology throughout coaching. (neptune-airflow 1.0.1)
On the roadmap 🔜
I’ll be again subsequent month with a reasonably large net app replace. Keep tuned.
Within the meantime, keep watch over our roadmap right here.