init_baseline Namespace Reference

Functions
def	emit (initializer_parameter_map)
def	run (initializer)
def	main ()

Variables
string	HEADER
string	FOOTER = "} // namespace expected_parameters"
string	PARAMETERS = "inline std::vector<std::vector<torch::Tensor>> {}() {{"
dictionary	INITIALIZERS

Detailed Description

Script to generate baseline values from PyTorch initialization algorithms

Variable Documentation

string init_baseline.HEADER

Initial value:

= """
#include <torch/types.h>

#include <vector>

namespace expected_parameters {
"""

Definition at line 6 of file init_baseline.py.

dictionary init_baseline.INITIALIZERS

Initial value:

= {
    "Xavier_Uniform": lambda w: torch.nn.init.xavier_uniform(w),
    "Xavier_Normal": lambda w: torch.nn.init.xavier_normal(w),
    "Kaiming_Normal": lambda w: torch.nn.init.kaiming_normal(w),
    "Kaiming_Uniform": lambda w: torch.nn.init.kaiming_uniform(w)
}

Definition at line 18 of file init_baseline.py.