wellcometrust / innovations_impact

This repository contains all the code needed to run the Innovations team's Lives Touched, Lives Improved model

Lives touched, lives improved model

This repository is for Innovations lives touched, lives improved model, which seeks to estimate the potential impact of interventions were they to be successfully developed. This model relies on assumptions abstract in order to model scenarios of potential impact - this results in highly uncertain estimates which is quantified.

How to setup:

This repository requires Python 3.6+.

Replicate the folder structure demonstrated here in you "My Documents" folder. All of the required files are found in the data folder.

The required packages are set out in requirements.txt.

If on a Mac, create and source a virtualenv with all dependencies via:

make
. build/virtualenv/bin/activate

Otherwise:

pip install --user -r requirements.txt 

You also need to sync this folder to OneDrive to be able to upload the model charts to the right place.

More detailed instructions on how to set up your computer and run the model can be found here.

How to run:

lives_touched_lives_improved.py is the main file to run the model from (it calls on the other modules in the repository).

It can be called from the command line and arguments passed to it to indicate the directories and file names where the required files (mentioned in the How to setup section above) can be found. Use the following command to see the possible inputs:

python lives_touched_lives_improved.py \
    --DATA_DIR data/ \
    --OUTPUTS_DIR outputs \
    --BACKUP_DIR backup/ \
    --SLIDES_DIR slides/ \
    --GRAPH_DIR graphs/

It can be also run from an IDE - at a minimum the file names and directory paths will need updating to be consistent with the folder structures on your machine - see more detailed instructions on that here.

To change the default behaviour of the lives_touched_lives_improved.py change the default terms in ANALYSIS_TYPE.

If ANALYSIS_TYPE is set 'run_all': False in lives_touched_lives_improved.py then you will have to input a valid id_code to run your chosen model. The id_codes come from LTLI_paramters.csv - check there to see which code you want to run. I would recommend having 'run_all': False as the default - as each model takes a couple of minutes to run on a standard Wellcome laptop.

How to model new projects:

All of the model parameters are contained in LTLI_parameters.csv. To model a new project then simply add a row to the csv with the new parameters.

Find instructions on how to input valid and sensible parameters here.

Note carefully:

1. There are not tests associated with all of the functions in this package.
2. There are input_checks which will throw errors if the datasets do not have the required columns, if in doubt then follow the format of the files handed over here (column names etc.).
3. The apply_exceptions.py module uses id_codes to make adjustments for a couple of projects modelled. These exception comments are tracked in the LTLI_outputs.csv - before making changes to an id_code (which I would not recommend generally) see if there is a relevant exception you are going to negate. The only exception I have applied is geographical limits on projects, a more general module could be written to apply this based on parameters, but hasn't been completed at time of writing.
4. Estimating the different models currently do not take advantage of memoisation - this means that coverage buckets simulated for one therapeutic are not reused for another and burden data simulated for a disease are not reused for a different intervention targetting the same disease, this would be a reasonably complex redevelopment but would cut down the run time of all the models dramatically. Another helpful addition would be to add in the potential to input a list of codes to run multiple models.