Install requires a python3 environment. I recommend using anaconda for linux, once anaconda is installed, I recommend making an environment to use for composed analysis, and ipython3 to use debugging capabilities:
conda update conda conda install pip conda create --name composed_analysis python=3 conda install ipython3 source activate composed_analysis
Once conda is installed, create your local composed library by checking out composed:
git clone https://github.com/Riedelbc/COMPOSED.git
Make a branch to do your local development. Replace my_composed_branch
with a meaningful name for your analysis project:
cd COMPOSED git checkout -b my_composed_branch
Install a local version of composed:
pip install -e .
Make a script in the COMPOSED/scripts/ folder, you can use one of the
existing ones as a template. If you are using a conda environment, you can't
just include #!/usr/bin/env python as the first line to run your script.
Instead, activate your python 3 environment, and then run:
source activate composed_analysis ipython3 --pdb %run filepath/scripts/your_script_name.py
This will make sure to use the local version of python that's located in your working python environment.
When you are finished running analyses, to return to the normal environment, hit control-d enter and then:
source deactivate composed_analysis
In your_script_name.py you need to set: the number of folds, classifier type, filepath to your csv, first/last column names of your csv
In your data.py in the "composed" dir, you need to set: prefixes of all variable classes on line 132 This is on the 'assert self.prefixes.issubset' line - if your feature type has the prefix 'VL_ROI' and 'Thk_ROI' then you would add 'VL' and 'Thk' - not ID or Group Also set the SVM baseline cutoff threshold 'cutoff_threshold' - you can also set this to 0
In your composed.py in the "composed" dir, you need to set: the c statistic to optimize your results This is on the 'clf = svm.LinearSVC' line
In your freesurfer.py script (or whatever it's called) there is a QC parameter in composed.py line 308 onwards That will prevent you from being able to run with more folds than you have data/group size for YOU NEED TO HAVE AT LEAST 3 DX/CN PEOPLE IN EACH FOLD! If not, reduce folds!
If/when it reaches an uncaught exception ipython3 will jump into debug mode. If/when that happens, here's the basics you need to know in order to figure out what went wrong.
- You can list out the surrounding lines of code if you type
l - You can travel up the call stack by typing
u - You can travel back down the call stack by typing
d - You can list out values of variables that are located at your current level of the call stack by typing the variable name.
- You quit the debugger by typing
q
Those are about the only commands that I use. If you want to learn more, check out the pdb docs.
Make sure to run:
git commit
Every once in a while in order to version control your work.
Do not save data inputs or data outputs into the repository, that will quickly
make the repository to big to use effectively. Instead, keep inputs and outputs
in the provided COMPOSED/inputs and COMPOSED/outputs directories. These
directories are listed in the COMPOSED/.gitignore file such that git will
not try and save files located in those directories.
If you make some interesting changes to composed tools and you want to share, feel free to make a pull request to share!