nnalgs's Introduction

Introduction

package for neural network algorithm training and tuning
uproot+lmdb+keras/keras-tuner
[email protected]

Installation

install anaconda
then restore the environments

conda env create -f CondaEnv.yml

(updated on 04/06/2020 using)
conda env export > CondaEnv.yml

git clone this repository

How to run it

# to train with static architecture
python NNAlgs Train NNAlgs/config/ConfDSNN.yaml
# to tune hyper-parameters
python NNAlgs Tune NNAlgs/config/ConfDSNN.yaml

How to add a new project

Layout

.
├── CondaEnv.yml
├── README.md
├── __main__.py
├── config
│   ├── ArchDSNN.yaml
│   ├── ArchLSTM.yaml
│   ├── ConfDSNN.yaml
│   ├── ConfLSTM.yaml
│   ├── DatasetBuilder.py
│   ├── GeneratorLoaders.py
│   └── __init__.py
├── data
│   ├── json
│   └── lmdb
├── nnalgs
│   ├── __init__.py
│   ├── algs
│   │   ├── DataGenerators.py
│   │   ├── LMDBCreators.py
│   │   ├── Models.py
│   │   └── __init__.py
│   ├── base
│   │   ├── BuilderDirector.py
│   │   ├── DataGenerator.py
│   │   ├── DatasetBuilder.py
│   │   ├── GeneratorLoader.py
│   │   ├── IDataset.py
│   │   ├── LMDBCreator.py
│   │   └── __init__.py
│   └── utils
│       ├── CallBack.py
│       ├── Common.py
│       ├── Enum.py
│       ├── LMDBGen.py
│       ├── Logger.py
│       ├── Math.py
│       └── __init__.py
└── nnexec
    ├── TrainDecayMode.py
    ├── TuneDecayMode.py
    └── __init__.py

TODOs

increase GPU usage [1]

[1] What does GPU-Util means in nvidia-smi

nnalgs's People

Contributors

Watchers

nnalgs's Issues

Validation Generator and Spliting

testing will be done in thor (more properly than doing it here but any plotting here is also good), don't worry
validation generator might need different batch size (larger)
so far 4:1 train/validation splitting, which is hardcoded, not good
in the future, a cross validation might also need to be considered..