NEMS
NEMS is the Neural Encoding Model System. It is helpful for fitting a mathematical model to time series data, plotting the model predictions, and comparing the predictive accuracy of multiple models. We use it to study computational models of how auditory stimuli are neurally encoded in mammalian brains, but it will probably work with your timeseries data as well.
Installation
git clone https://github.com/lbhb/nems
# Please make sure that the nems repo is in your path. This works for linux:
echo "export PYTHONPATH=\"\$PYTHONPATH:`pwd`/nems\"" >> ~/.bashrc
source ~/.bashrc
We have found that Python 3 distributions compiled with the Intel MKL libraries are about twice as fast as the default Python implementations that come installed on many linux machines. Please see our basic conda installation instructions if you would like to set up a python environment like the way that we do.
NOTE: Regardless of which Python distribution you choose to use, NEMS is designed to use Python 3. Backwards compatibility with Python 2 is untested and unsupported.
Finally, if for some reason they are not installed already, you will also need to use pip to install several libraries that NEMS uses:
pip install requests numpy scipy matplotlib pandas
Alternatively, you may use pip to automatically install all libraries specified as required by NEMS.
# Where 'NEMS' refers to the name of the directory containing the installation
pip install -e NEMS
Your First Model Fit
You may test if everything is working by telling NEMS to download some sample auditory stimulus-response data, use a simple linear-nonlinear model (which should taking about 2 minutes to fit), and then save the results locally:
cd nems/scripts
INPUT_URI=https://s3-us-west-2.amazonaws.com/nemspublic/sample_data/TAR010c-18-1.tgz
MODELKEYWORDS="wc18x1_lvl1_fir15x1_dexp1"
DESTINATION=~/nems_results/
python fit_model.py $INPUT_URI $MODELKEYWORDS $DESTINATION
Now you should go in ~/nems_results and find the subdirectory where the plot PNG files, modelspecs, and other files were saved. As I hope you can see, the $INPUT_URI and $DESTINATION variables may either be local paths, or HTTP URLs. For example, you might want to download that file for later, offline use in model fitting:
wget https://s3-us-west-2.amazonaws.com/nemspublic/sample_data/TAR010c-18-1.tgz -P /tmp
INPUT_URI=/tmp/TAR010c-18-1.tgz
MODELKEYWORDS="wc18x1_lvl1_fir15x1_dexp1"
DESTINATION=~/nems_results/
python fit_model.py $INPUT_URI $MODELKEYWORDS $DESTINATION
In our laboratory, we have a NEMS_DB server that we save our fit models to, so the INPUT_URI is usually an URL like http://ourserver.edu/recordings/... and the destinations are an URL like http://ourserver.edu/results. If you will be saving, searching, and sharing lots of model fits with other people in your lab, you may want to consider installing NEMS_DB as well.
Table of Contents
This documentation is a work in progress as of March 9, 2018.
- Quick Start
- Organizing your Data
- Organizing your Models
- Fitting your Models
- Detailed Guides
- Architectural Diagram
- Creating your own modules
- Comparing your models with others
- Sharing modules, models, and data with others
- Contributing to NEMS
- Other