Some test to partially shared layer between network
I made this repository to try different way to shared weight in different configuration :
- siamese network
- common parts (We want a model able to predict Age and Gender but the database don't contain both attributs)
In 2017, the gluon API was still new that's why I try to use the Module API. Today, I advice to use the gluon API, it's more flexible and don't require hack to achieve shared weight.
.
├── demo_shared_partial_with_super_symbol_v1.py not working: Example with Shared_module argument from mx.mod.bind
├── demo_shared_partial_with_super_symbol_v2.py not working: Example with Shared_module argument from mx.mod.bind
├── demo_shared_with_weight_variable.py not working: Ewample with shared weight
├── demo_with_gluon_one_network.py Example with Gluon without shared weight
├── demo_shared_full_net.py Example with Shared_module argument from mx.mod.bind
├── demo_shared_with_weight_copy_each_time.py Example with weight transfert each time
├── demo_with_gluon.py Example with Sequential gluon API
├── demo_with_gluon_hybrid.py Example with HybridSequential gluon API
├── demo_with_gluon_inside_param.py Example with condition in Block forward definition
├── demo_with_gluon_siamese.py Example of siamese network with gluon
└── README.md
