Some codes and results about our paper, here is the structure of our directory:
-
/src: High level interface of out implementation of model generator across frameworks by JSON
- configuration: Some configurations and basic settings
- dataset: Dataset across frameworks API
- grammar: Grammar checker for model generation
- interpreter: Model Interpreter for model generating across frameworks by JSON
- metrics: Metrics across frameworks API
- optimizers: Optimizers across frameworks API
- test_process: Test process across frameworks API
- train_process: Train process across frameworks API
- model_json.py: Entrance implementation for model generation
- *.json: Examples for JSON files
-
/torchsummary: Open source package for the function of summary() for PyTorch
-
/trials: Codes about Gandalf experiments
-
classical: An implementation of Audee with models of VGG16, Lenet-5 and Resnet-20
-
envs: The test environment for our testing process, currently, only the environment of CNN2d is supported
-
dqn: The main implementation of out method and random method
-
results: Some results of our experiments
-
Through our method, we have found some serious faults for Tensorflow and PyTorch, and we have submitted them to the issues as follow, most of them are accepted: