weight_initializer

• type: xavier normal (default) / xavier uniform, normal, constant, kaiming normal, kaiming uniform, he normal, he uniform, orthogonal, sparse

---

Weight initialization is used to define the initial values for the parameters in Neural Network models prior to training the models on dataset. SIMPLE-NN supports several weight_initializer and the usage of weight_initializer is as below.

# input.yaml
neural_network:
    weight_initializer:
        type: xavier normal
        params: