MLflow v1.13 Release Notes
Release Date: 2020-12-22 // over 3 years ago-
MLflow 1.13 includes several major features and improvements:
๐ Features:
๐ New fluent APIs for logging in-memory objects as artifacts:
- โ Add
mlflow.log_text
which logs text as an artifact (#3678, @harupy) - โ Add
mlflow.log_dict
which logs a dictionary as an artifact (#3685, @harupy) - โ Add
mlflow.log_figure
which logs a figure object as an artifact (#3707, @harupy) - โ Add
mlflow.log_image
which logs an image object as an artifact (#3728, @harupy)
โก๏ธ UI updates / fixes (#3867, @smurching):
- โ Add model version link in compact experiment table view
- โ Add logged/registered model links in experiment runs page view
- โจ Enhance artifact viewer for MLflow models
- ๐ป Model registry UI settings are now persisted across browser sessions
- โ Add model version
description
field to model version table
Autologging enhancements:
- ๐ Improve robustness of autologging integrations to exceptions (#3682, #3815, dbczumar; #3860, @mohamad-arabi; #3854, #3855, #3861, @harupy)
- โ Add
disable
configuration option for autologging (#3682, #3815, dbczumar; #3838, @mohamad-arabi; #3854, #3855, #3861, @harupy) - โ Add
exclusive
configuration option for autologging (#3851, @apurva-koti; #3869, @dbczumar) - โ Add
log_models
configuration option for autologging (#3663, @mohamad-arabi) - Set tags on autologged runs for easy identification (and add tags to start_run) (#3847, @dbczumar)
More features and improvements:
- ๐ Allow Keras models to be saved with
SavedModel
format (#3552, @skylarbpayne) - โ Add support for
statsmodels
flavor (#3304, @olbapjose) - โ Add support for nested-run in mlflow R client (#3765, @yitao-li)
- ๐ Deploying a model using
mlflow.azureml.deploy
now integrates better with the AzureML tracking/registry. (#3419, @trangevi) - โก๏ธ Update schema enforcement to handle integers with missing values (#3798, @tomasatdatabricks)
๐ Bug fixes and documentation updates:
- When running an MLflow Project on Databricks, the version of MLflow installed on the Databricks cluster will now match the version used to run the Project (#3880, @FlorisHoogenboom)
- ๐ Fix bug where metrics are not logged for single-epoch
tf.keras
training sessions (#3853, @dbczumar) - ๐ฒ Reject boolean types when logging MLflow metrics (#3822, @HCoban)
- ๐ Fix alignment of Keras /
tf.Keras
metric history entries wheninitial_epoch
is different from zero. (#3575, @garciparedes) - ๐ Fix bugs in autologging integrations for newer versions of TensorFlow and Keras (#3735, @dbczumar)
- โฌ๏ธ Drop global
filterwwarnings
module at import time (#3621, @jogo) - ๐ Fix bug that caused preexisting Python loggers to be disabled when using MLflow with the SQLAlchemyStore (#3653, @arthury1n)
- ๐ Fix
h5py
library incompatibility for exported Keras models (#3667, @tomasatdatabricks)
โก๏ธ Small changes, bug fixes and doc updates (#3887, #3882, #3845, #3833, #3830, #3828, #3826, #3825, #3800, #3809, #3807, #3786, #3794, #3731, #3776, #3760, #3771, #3754, #3750, #3749, #3747, #3736, #3701, #3699, #3698, #3658, #3675, @harupy; #3723, @mohamad-arabi; #3650, #3655, @shrinath-suresh; #3850, #3753, #3725, @dmatrix; ##3867, #3670, #3664, @smurching; #3681, @sueann; #3619, @andrewnitu; #3837, @javierluraschi; #3721, @szczeles; #3653, @arthury1n; #3883, #3874, #3870, #3877, #3878, #3815, #3859, #3844, #3703, @dbczumar; #3768, @wentinghu; #3784, @HCoban; #3643, #3649, @arjundc-db; #3864, @AveshCSingh, #3756, @yitao-li)
- โ Add