Locations

Folders

• LinearA_trained contains a pretrained model.
• FTIR_to_electrolyte_composition contains code
  • models.py defines how the dataset is stored.
  • management/commands/LinearA_run.py is where the code functionality is defined

Files

• .gitattributes can be ignored
• .gitignore can be ignored
• db.sqlite3 contains the dataset
• LICENSE contains the license
• manage.py is the entry point for the program.
• requirements.txt contains the requirements. (excluding python 3.6)

Source Code

see FTIR_to_electrolyte/management/commands/LinearA_run.py

Usage

There are many possible ways to call the code:

1. After having put some measurement files in a directory, the model can be called by specifying this directory. It will return an excel file with the predicted mass ratios, as well as a directory with a graph of the reconstructed spectra together with the original data for each input file (the extension of the files will be changed from .asp to .png). This is the run_on_directory option. For convenience, an excel sheet with the numerical data for measured and reconstructed spectra is outputted as well so the user can make their own graphs (e.g. for publication).
2. The training can be run on the calibration dataset, and thus the model can be updated. This is the train_on_all_data option.
3. Cross-validation studies of the model can be run to evaluate it on the calibration dataset. This is the cross_validation option.
4. The figures in this paper can be reproduced. This is the paper_figures option.
5. There is a functionality to create a dataset, allowing the extension of the dataset, but it might require some modification to adapt to a different lab's workflow. This is the create_dataset option. It is not documented because the way to specify the known weight ratios will have to change for new data.

here are some examples of calling the code:

• "python manage.py LinearA_run --mode=run_on_directory --logdir=LinearA_trained --input_dir=test_input_data --output_dir=Output" will run the trained model stored in folder 'LinearA_trained' on all the .asp files found in folder 'test_input_data', putting the output in folder 'Output'

• "python manage.py LinearA_run --mode=train_on_all_data --logdir=LinearA_trained" will train a model on the complete dataset and store the trained model in folder 'LinearA_trained'

• "python manage.py LinearA_run --mode=cross_validation --test_ratios=0.3 --cross_validation_dir=Cross_LinearA" will run a single training procedure with 30 percent of the data in the test set, storing the result in folder "Cross_LinearA".

• "python manage.py LinearA_run --mode=paper_figures --cross_validation_dir=Cross_LinearA" will produce the figures from the paper for the cross validation runs stored in folder "Cross_LinearA".

Requirements

see requirements.txt install the requirements by running on the command line "pip install -r requirements.txt"

Understanding the database

An in-depth guide to the dataset and how to access it in python is written at the begining of the file FTIR_to_electrolyte_composition/models.py

Understanding the model that maps FTIR spectra to the mass ratios of various components

This README contains valuable information about how to call the program to accomplish various operations (train, run on a directory, do cross-validation, etc..) For more in-depth documentation of how the model is implemented, see FTIR_to_electrolyte_composition/management/commands/LinearA_run.py