v0.8.322

• flake8/linter fixes
• adding a df_store script to pull and push dataframes from one AWS env to another
• flake8/linter fixes
• simplify the endpoint autoscaling utils

v0.8.321

• putting in a conditional so that only regression models calculate uq_metrics()

v0.8.320

• bumping version of workbench for ml_pipeline image

v0.8.318

• put in some NaN guards when splitting on scaffolds or butina
• add a script for smiles/NaN checks
• flake8/linter fixes
• small change to have multi-target models generate same outputs as single-target
• flake8/linter fixes
• adding some pytorch experiments
• setting restore_best_weights to False by default
• flake8/linter fixes
• adding threading to the AllChem.MMFFOptimizeMoleculeConfs call
• adding some reference compounds and profiling for 3D features
• blog grid spacing update

v0.8.317

• flake8/linter fixes
• changing/optimizing some of the 3D Feature calculations
• bumping version of workbench for ml_pipeline image

v0.8.316

• making different varients of the pytorch models
• flake8/linter fixes
• changing the default splits to be scaffold
• adding some confidence tests
• removing the df[prediction_std] overwrite as the templates already compute this internally
• some cleanup

v0.8.315

• flake8/linter fixes
• changes to InferenceCache to handle typing across chunks more robustly
• flake8/linter fixes
• converting boltzman vars/references to full

v0.8.314

• some tweaks to batch/chunk logic

v0.8.313

• bumping version of workbench-bridges
• adding a compaction admin script
• flake8/linter fixes
• trying to simplify logic around max_instances, chucks, and batches
• new logic for inference cache to manage concurrency issues
• adding an admin script that tests inference cache concurrency
• adding a preflight check to make sure the account has the quota for max_instances of that specific instance type
• putting in max_in_flight logic
• changes to the auto-scaling for endpoints
• moving the ParameterStore implementation over to workbench-bridges
• refactoring how feature_lists() are computed (mostly internal now)
• flake8/linter fixes
• adding a FeatureEndpoint() class that has a feature_list() method
• flake8/linter fixes
• making the new async args more visible through API, also moving common code into the feature_endpoint utils
• adding support for min_instances, max_instances, auto_scaling_mode, parameters when setting up an AsyncEndpoint
• flake8/linter fixes
• put in a max_instances arg that's used for AsyncEndpoints
• flake8/linter fixes
• adding a register_features() method so that it can be used by downstream consumers
• adding a register_features() method so that it can be used by downstream consumers
• adding a register_features() method to EndpointCore, so that it can be used by downstream consumers

v0.8.312

• logic fix for detecting async endpoints
• flake8/linter fixes
• renaming all the feature endpoints
• adding a reference set for flexible compounds + test