https://github.com/eddytheco/fisher_app
https://github.com/eddytheco/fisher_app
Last synced: 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/eddytheco/fisher_app
- Owner: EddyTheCo
- Created: 2022-11-02T16:31:41.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2022-12-01T11:50:35.000Z (over 2 years ago)
- Last Synced: 2025-02-07T08:34:39.095Z (4 months ago)
- Language: CMake
- Size: 1.11 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Effective Dimension and Fisher Spectrum calculation
This repository produce an example application to calculate the effective dimension and Fisher spectrum of a statistical Torch module.
It shows how to use the [Fisher Matrix library](https://github.com/EddyTheCo/Fisher_Matrix) for this purposes.
It uses [yaml](https://github.com/jbeder/yaml-cpp) [input files](INPUT) for the configuration of the application.
## Dependencies
* libtorch from [Pytorch](https://pytorch.org/)
## CMake variables for configuration
|name|values|description|
|----|------|-----------|
|MODEL|DNN_abbas,EQM_abbas,QNN_abbas ..|The torch module to use as a statistical model and calculate the Fisher Information Matrix. One can create custom modules and add it to the namespace custom_models.|
## Install and Execute
### Build and install
```
git clone [email protected]:EddyTheCo/Fisher_app.git Fisher_app
cd Fisher_app
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=install -DMODEL=DNN_abbas -DCUSTOM_MODULES="DNN_abbas" ../
cmake --build . --target install -- -j4
```
### Execute
```
cd install
./bin/fisher_app INPUT/dnn_abbas_config.yaml
```
### Results
The normalized spectrum of the calculated Fisher matrices is written in the file Spectrum.txt.
The Effective dimension results are written to the file Effect_dime.txt.
To reproduce the results from the paper one could use the gnuplot scripts like:
```
gnuplot gnuplot/effectiveD.gnu
gnuplot gnuplot/eig_distr.gnu
```
When trying to reproduce the results of the paper the obtained results were
**Quantum neural network**
**Easy quantum model**
**Classical neural network**
