Deep Learning Complexity Measure Experiments
This is the code that compares various complexity measures for deep neural networks, recently proposed in the literature.
Evaluation relies on three tools:
- src/eval.py -- runs comparison experiments and stores them as pickled dicts.
- src/collect_results.py -- collects results produced by eval.py into a sqlite database.
- src/results.py -- plotting and table formatting tools that read from sqlite database.
bin/mnist_demo.sh showcases an examples of evaluation workflow.