This repo contains different part of Marker Intensity Research, based on HTAN data files.
It includes:
- Jupyter notebooks
- A linear regression model to calculate LR baseline data
- Different auto encoders to train models
- A plotting package to combine the results of linear regression and different auto encoders
Run the following to start the jupyter notebook:
$ virtualenv .venv && source .venv/bin/activate && python -m pip install -r requirements.txt && python -m ipykernel install --user --name=.venv
$ jupyter labextension install [email protected]
$ jupyter-lab- Create a virtual environment
pip install -r requirements.txtwill install all required packages
MLFlow is being used for experiment tracking.
Although a results folder is being used locally, all data is being stored using the backend of MLFlow.
Do NOT rely on the local results folder, it is only temporary.
python VAE/main.py model --file data/HTA9-3_Bx2_HMS_Tumor_quant.csv
| Long | Short | Description | Required |
|---|---|---|---|
| --file | x | The file to use for training the models | ✓ |
| --run | -r | The name of the run | ✓ |
| --experiment | -e | The name of the associated experiment. If no experiment with this name exists and new one will be created. Default: Default experiment |
|
| --morph | x | Should morphological features be included for training and evaluation? Default: true | |
| --mode | x | Possible values: none, ae, vae. If ae or vae is selected only the respective model is being trained. No comparison will be done. Default: none. |
|
| --tracking_url | -t | The tracking url for the mlflow tracking server. Points to localhost as default. Please specify a valid url. Example: http:127.0.0.1:5000 |
A tool to explore the latent space of a VAE is included.
The tool is using Streamlit and MLFlow to provide interactive as well as tracking functionalities.