Practical overview
When it fits
Use it when you need research scripts and experiment examples for ICNNs, especially if your work involves classification, face completion or Gym-based reinforcement learning.
Who it is for
- Researchers reproducing or adapting ICNN experiments.
- Developers exploring convex neural-network methods across classification, image completion or reinforcement learning.
Documented uses
- Reproduce synthetic classification experiments on non-convex decision boundaries.
- Train and plot multi-label ICNN experiments using the included Bibsonomy loader.
- Train and inspect Olivetti face-completion experiments.
- Run reinforcement-learning experiments with ICNN, DDPG or NAF model choices.
Requirements & limits
The repository is public and its source code is licensed under Apache-2.0.
- The general setup lists Python, NumPy and TensorFlow, with TensorFlow r10 cited; Gym and MuJoCo are additional dependencies for reinforcement-learning work.
- The synthetic classification run script is described as running all experiments on four GPUs.
- The README says multi-label experiments are slightly inconsistent with the paper.
- The bundle-entropy utility is described as in progress, with numerical issues still being fixed.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run a single ICNN reinforcement-learning experiment
Launch the README's documented InvertedPendulum experiment and identify its stated output files.
Before you start
- Python, NumPy and TensorFlow; the setup notes cite TensorFlow r10.
- OpenAI Gym and MuJoCo for reinforcement-learning experiments.
- The repository's RL directory.
Steps
From the RL directory, run the command below.
python src/main.py --model ICNN --env InvertedPendulum-v1 --outdir output --total 100000 --train 100 --test 1 --tfseed 0 --npseed 0 --gymseed 0Use the README's parameter help command, the command below, to view available parameters.
python main.py -h
Expected outcome
The README says TensorBoard summaries are on by default and include average Q value, loss and average reward for each training minibatch. Testing total rewards are logged to log.txt.
Guide limitations
- Gym and MuJoCo are listed as requirements for the RL experiments.
- The README describes settings for reproducing experiments separately but does not provide them in this single-experiment example.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- locuslab/icnn · Official homepage title · retained in its source language