Optimization - Multifactor Elasticity 1.1
- Ondřej Tesař
This document summarizes major improvements and fixes introduced in the Accelerate Multifactor Elasticity Optimization package release version.
Version | 1.1.0 |
|---|---|
Release Date | Jun 15, 2023 |
Table of contents:
New Features and Improvements
Description | ID |
|---|---|
There is a new bar chart showing the feature importance and a new table with important metrics from training and test data sets. This helps to explain which features are important to the trained model to assist with feature selection and to increase explainability of the model. | PFPCS-6772 |
If not yet configured for the partition, Python Engine is enabled automatically during accelerator deployment. | PFPCS-6943 |
It is possible now to define the price range of the predicted elasticity curves. | PFPCS-6953 |
The maximum forecast length has been extended to 15 time periods. | PFPCS-6954 |
Fixed Issues
Bug Description | ID |
|---|---|
In the Definition step, when you leave "Additional categorical features" or "Additional numerical features" empty (even if they are optional), you get an error during the Training step calculation. | PFPCS-6770 |
Fixed an issue where the default number of estimators used for model training was too low, limiting performance on large datasets. Model results are now reproducible for the same data and inputs. | PFPCS-6799 |
Some user inputs (number of lags, diffs, rolling window) in elasticity calculation were missing, as they were hard-coded and not using the user inputs. Those inputs are now used correctly in the calculation. | PFPCS-6958 |