Releases: volkamerlab/maxsmi
Releases · volkamerlab/maxsmi
v1.1.0
v1.0.0
This release is the one which was used to generate the results for the maxsmi paper, see https://doi.org/10.1016/j.ailsci.2021.100014 .
Submission
This release addresses a few key points:
- Updated README: more content: project description, folder structure, examples, ...
- Analysis of results in notebooks (see
maxsmi/results_analysis/
). - Prediction tool: save the weights of the Maxsmi models for out-of-sample prediction (see
maxsmi/prediction_models/
). - Documentation update: example on user prediction and more detailed examples.
Simulations
Simulation runs
This version is the one that was used to run all the simulations (on a cluster):
- The three datasets
- ESOL
- free_solv
- lipophilicity
- The three machine learning models
- CONV1D
- CONV2D
- RNN
- The five augmentation strategies:
- no augmentation
- augmentation without duplication
- augmentation with duplication
- augmentation with reduced duplication
- augmentation maximum estimation
- Finer and coarser grid length:
- 1 : 20 : 1
- 20 : 100 : 10