GAMCR in practice: A brief overview

Typical File Structure for GAMCR

GAMCR/
├── experiments/
│   ├── check_data.ipynb
│   ├── save_data_batch.py
│   ├── train_models.py
│   ├── compute_statistics.py
│   └── mysite/
│       ├── results/
│       │   ├── detailedresults/
│       │   ├── groups.csv
│       │   ├── NRF_RRD.csv
│       │   └── streamflow.csv
│       ├── data/
│       │   ├── transfer.npy
│       │   ├── lst_transfer.npy
│       │   ├── y_0.npy
│       │   └── ...
│       ├── data_mysite.txt
│       └── mysite_best_model.pkl

Folder Structure for Using the GAMCR Package

To properly use the GAMCR package, you should have a folder for each site with the following structure:

• You can name the folder as per your preference. For example: mysite.

• Inside this folder, you should have a file named data_{mysite}.txt. This file can be created by running the notebook check_data.ipynb.

• Two subfolders will be created and used by GAMCR:

  • ``data/``: This subfolder will store the preprocessed data, created when running the script save_data_batch.py.

    • For simulated data:

      • If the true transfer functions have been computed, they are saved in a matrix in the file transfer.npy.

      • The indices corresponding to each precipitation event used to compute the true transfer functions are saved in the file lst_transfer.npy.

  • ``results/``: This subfolder will store various statistics on the results of a trained model, created when running the script compute_statistics.py. Specifically, the following files will be generated:

    • groups.csv: Describes the precipitation and/or antecedent wetness ranges for each group of data points used to compute averages.

    • NRF_RRD.csv: Contains NRF, RRD, and weighted average RRD computed for the different groups of data points.

    • streamflow.csv: Includes observed and predicted streamflow time series.

• GAMCR will also save the models you train for that site in this folder. By running the script train_models.py, a file named {mysite}_best_model.pkl will be generated.