v0.8.304

• flake8/linter fixes
• some simplification of the inference cache internal logic
• flake8/linter fixes
• new InferenceCache class that wraps endpoints and caches results
• flake8/linter fixes
• adding a conformer timeout mechanism
• don't print out the metrics when they are empty
• updating 3D feature endpoint docs
• changing the complexity logic and adding more details on the SMARTs patterns
• changing the number of threads option for the conformation
• putting in a bit of logic to not recreate the model
• changing the batch_size and bigger image
• updating the 3D docs
• flake8/linter fixes
• add a 3D Endpoint test
• flake8/linter fixes
• a bunch of changes for the 3D descriptor endpoint and supporting utils/tests
• flake8/linter fixes
• updating the complexity check and ripping out the timeout that doesn't work with C++ code/optimizer
• flake8/linter fixes
• adding corner case tests
• adding a UFFHasAllMoleculeParams() check before sending to conformer
• adding 3D molecular descriptors in the overview page
• flake8/linter fixes
• adding a blog about our 3D descriptors endpoint
• a bunch of updates to our 3D descriptors calculations
• 3D endpoint tuning: bump serverless memory to 6144MB and reduce batch size to 10
• SageMaker v3 fixes: OutputDataConfig compression_type and training_harness auto-copy
• add a readme that show how to deploy
• adding the training_harness copy for custom model scripts
• adding OutputDataConfig args for our model training parameters
• flake8/linter fixes
• adding a complexity check for the 3D Features Calculation to address slow endpoint times
• bumping versions of workbench for ml_pipeline image

v0.8.303

• fix the imports for our monkey patch
• monkey path a bug in Sagemaker v3 DataCaptureConfigSummary.kms_key_id

v0.8.302

• changing start_date to after_date to be more explicit
• a refactoring of how the temporal split logic works to support dates

v0.8.301

• bumping the version of workbench-bridges
• removing the refresh session code.. wasn't the bug/issue we were looking for

v0.8.300

• simplify the disabling of the sagemaker rich logging
• adding functionality to reset our aws session
• bumping versions of workbench for ml_pipeline image

v0.8.299

• changes to the combine_multi_task_data functionality
• increasing the batch time out to 24 hours
• flake8/linter fixes
• adding a public data test
• flake8/linter fixes
• adding a sleep after the sample_weights table delete
• adding some multi task utils
• just some cleanup
• adding a _disable_sagemaker_rich_logging() method to disable Sagemaker v3 terrible 'rich' logging

v0.8.298

• flake8/linter fixes
• adding a spawn test for FeatureGroup ingestion
• adding a spawn test for FeatureGroup ingestion
• putting in logic to have normal amount of workers/processsors on linux but fall back to single thread on Mac (spawn bug in sagemaker)
• flake8/linter fixes
• switching over to using the new PublicData class

v0.8.297

• installing workbench and workbench-bridges in a separate layer
• put in a fix for sample rows

v0.8.296

• flake8/linter fixes
• adding public data outputs to the git ignore
• adding functionality to pull public data
• adding an import alias to workbench-bridges PublicData
• bumping the version of workbench-bridges
• adding a known counterions list that used as part of the fragment processing
• add handling for ReadTimeoutError for Endpoints under load
• adding a LaplacianSmoothness class
• adding a LaplacianSmoothness class
• small logic change to neighbor_median
• adding a python link in our pytorch_chem training Dockerfile

v0.8.295

• fix path for ml_pipeline image creation
• rename TestDataGenerator to SyntheticDataGenerator
• setting 6 hour max training time
• flake8/linter fixes
• moving over to the new training harness that works with Sagemaker V3
• removing include_code_and_meta_for_inference() from model templates
• removing include_code_and_meta_for_inference() from model templates
• some code coverage setup
• updating the workbench-bridges version to 0.2.0
• small readme change
• a large refactor of the sagemaker images for training and inference