This package was built on the Keras framework and easily allows using the PlexusNet architecture for research projects.
- To Install:
pip3 install git+https://github.com/oeminaga/PlexusNet.git - Please install tensorflow, tensorflow_addons and tensorflow_probability before using this package or use requirements.txt before starting with install.
pip install -r requirements.txt
- For reproducibility, please add the function
import plexusnet
seed_everything()
- An example code to use the package:
from plexusnet.architecture import PlexusNet
model=PlexusNet(depth=2, length=3, junction=3, n_class=2).model
model.compile(optimizer="adam", metrics=["acc"], loss="categorical_crossentropy")
model.fit(X,Y)
- If you want to load a PlexusNet model for prediction:
from plexusnet.architecture import LoadModel
model=LoadModel("your_model.h5")
model.predict(X)
- For NAS (only one epoch for each model configuration), example codes are provided:
-- Grid search
(https://github.com/oeminaga/PlexusNet/blob/master/Example_Cifar_PleuxsNET-GridSearch.ipynb)
-- Bayesian Optimization
(https://github.com/oeminaga/PlexusNet/blob/master/Example_Cifar_PleuxsNET_BayesianOptimization.ipynb)
-- Hyperband
(https://github.com/oeminaga/PlexusNet/blob/master/Example_Cifar_PleuxsNET_Hyperband.ipynb)
-- Random search
(https://github.com/oeminaga/PlexusNet/blob/master/Example_Cifar_PleuxsNET_RandomSearch.ipynb)
Please exclude h5 in the filename if you want to save the model during training with TF version >2.8 to avoid the model saving errors!
Please exclude the option "save weight only" to avoid errors when you wanted to load the model. Instead, save the whole model during model training.
Bayesian block is now added (bayesian_inception).
Please cite, when you use this package:
O. Eminaga, M. Abbas, J. Shen, M. Laurie, J.D. Brooks, J.C. Liao, D.L.
Rubin, PlexusNet: A neural network architectural concept for medical image classification, Computers in
Biology and Medicine (2023),doi: https://doi.org/10.1016/j.compbiomed.2023.106594.