Integrating domain adaptation

[MWesselkamp]

We use ENmlp, OFmlp and ftmlp as before. Create new files:

1. simulations.py
2. pretraining.py
3. evalmlpDA.py

Pretraining will be called from evalmlpDA.py with pretraining settings.

[MWesselkamp]

pretraining.py

Changes required in utils.loaddata():

• enable to load simulations for pretraining

• [ ] Set data splits to the number of data splits used for evaluation on observations

[MWesselkamp]

simulations.py

Move Preles simulations from R to python. DONE.

• [ ] Decide on parameter space dimensions (so far 5)
• [ ] Adjust upper an lower boundaries in parameterRanges.csv to Carstens new boundaries.
• [ ] How many simulations?
• [ ] Simulate whole year or shorter sequences for the same parameter values?

Generating simulations by submitting generateDAsimulations.sh with adjusted paramters.

[MWesselkamp]

evalmlpDA.py

Finetuning function or argument required (in training.py) to enable layer freezing.

• [ ] Enable feature extraction in training.py (domain_adaptation = 2)
• [ ] Reset epochs to orginal iterations