analysis: build 3-view moma

[Piotr1619]

The PR contains a jupyter notebook run_triple_view_moma.ipynb to run the whole simulation (training & validation) for 3-view MOMA model that now includes three inputs: transcriptomics, fluxomics, proteomics

[tdsone]

• the proteomics model loss makes me really optimistic that this could work - good job!
  

• make sure to log the loss function and check that your model is learning (read up on how to interpret loss curves)

• there is something broken with the training of the 3-view model

• restructure your code into separate python files (a useful split is: model.py (model architecture), train.py (train + eval), preprocessing.py (data prerpocessing code)

• you could also think about having a separate folder for the proteomics model and the 3 view model

• jupyter notebooks are a bit hard to review as you cannot track changes properly and they are inherently messy

looking forward to the next iteration

[Piotr1619]

Thanks for your comments! Here is what I improved so far:
I split the code into the following architecture: moma/model.py (here, we keep all the models used in moma). Then for our analysis, I created ralser_moma folder where we have ralser_preprocessing.py, ralser_train.py, and ralser_main.py.
To run the script type: python bench/models/moma/ralser_moma/ralser_main.py
For the input data, you need: data/models/moma/yeast5k_impute_wide.csv and data/tasks/task3/yeast5k_growthrates_byORF.csv

Currently, the script saves automatically the loss plot into data/models/moma/proteomics_model_loss.png and weights into data/models/moma/proteomics_ralser.weights.h5

I'm happy with any more suggestions 👍

[tdsone]

I really like this PR! Was much easier to review and spot things. You can merge but make an issue or smth to add instructions about setting up the datasets in the README. Really looking forward to the final weeks. If you want to flex a bit: make a gallery of the MSE/loss curves and show them as before and after images in the subgroup meeting!