This repository was build using a virtual environment unsing venv. The requirements file to generate this environment is saved as requirements.txt
| model file | description |
|---|---|
| lib/FTS_Mo_4 | default FTS model with logit scaling |
| lib/FTS_Mo_5 | Verfloet model with logit scaling |
These two model files are used to generate all results depicted in the publication.
The model refinement using bDOE was done using the script HSA_007_AE.py which accesses the available data and plans the next experiments.
Evaluation was run using HSA_007.ipynb to reproduce the figures from the publication. HSA_009.py and HSA_009.ipynb were used to calculate the information gain of the initial bDOE.
Literature comparison to Vervloet et al. was done using the script HSA_010.ipynb.
This data was used to generate Fig. 2 in the manuscript:
The experimental data sets are saved.
| file | data set |
|---|---|
| data/SHB_001.csv | training data |
| data/SHB_002.csv | validation data |
All inference files are recorded and saved under their run name associated to the training script:
| file | model | ID |
|---|---|---|
| HSA_005 | lib/FTS_Mo_1 (old) | data/rude-bicycle |
| HSA_006 | lib/FTS_Mo_4 (no constraints) | data/profound-aim |
| HSA_007 | lib/FTS_Mo_4 (default) | data/scruffy-decibel |
Inference file for the Vervloet model is not saved due to low calculation times. Inference method is in the file.
The bDOE in HSA_007 (data/scruffy-decibel) suggested four additional experiments which are supplied as input to the algorithm in data/scruffy-decibel/data/{idx}.csv.