pytorch-initialize-module-from-yaml

• Python function which returns pytorch.nn.module according to configuration yaml file

Rules

• def initialize_custom_model(spec: dict) -> pytorch.nn.Module
• 'spec' must have form of below. Be careful of RNN input/output!

initialization: bool
layers: [layer_element]

• 'layer_element' must have some forms of:

layer: 'linear'
spec: [INPUT_SIZE, OUTPUT_SIZE]
activation: activation

layer: 'conv2d'
spec: [IN_CHANNEL, OUT_CHANNEL, KERNEL_SIZE, STRIDE, PADDING]
activation: activation

layer: 'rnn'
spec: [INPUT_SIZE, HIDDEN_SIZE, NUM_LAYERS]
activation: activation

layer: 'lstm'
spec: [INPUT_SIZE, HIDDEN_SIZE, NUM_LAYERS]
activation: activation

layer: 'flatten'

• 'activation' must have form of one of:

- relu
- elu
- softmax
- none

Dependency

• Numpy
• PyTorch
• PyYaml

Run

python3 main.py