nnfw_train_info Struct Reference

Training information to prepare training. More...

#include <onert-micro.h>

Collaboration diagram for nnfw_train_info:

Data Fields
float	learning_rate = 0.001f
uint32_t	batch_size = 1
nnfw_loss_info	loss_info
NNFW_TRAIN_OPTIMIZER	opt = NNFW_TRAIN_OPTIMIZER_ADAM
uint32_t	num_trainble_ops = 0
nnfw_adam_option	adam_opt
int32_t	num_of_trainable_ops = NNFW_TRAIN_TRAINABLE_NONE

Detailed Description

Training information to prepare training.

Definition at line 186 of file onert-micro.h.

Field Documentation

adam_opt

nnfw_adam_option nnfw_train_info::adam_opt

Initial value:

{.beta = 0.9f,
 
                            .beta2 = 0.999f,
                            .epsilon = 1e-7f}

Definition at line 200 of file onert-micro.h.

                           {.beta = 0.9f,
 
                            .beta2 = 0.999f,
                            .epsilon = 1e-7f};

batch_size

uint32_t nnfw_train_info::batch_size = 1

Batch size

Definition at line 191 of file onert-micro.h.

learning_rate

float nnfw_train_info::learning_rate = 0.001f

Learning rate

Definition at line 189 of file onert-micro.h.

loss_info

nnfw_loss_info nnfw_train_info::loss_info

Initial value:

{.loss = NNFW_TRAIN_LOSS_CATEGORICAL_CROSSENTROPY,
                           .reduction_type = NNFW_TRAIN_LOSS_REDUCTION_SUM_OVER_BATCH_SIZE}

loss info

loss info Note that you don't need to worry about whether the model you use does not include softmax when you try to use NNFW_TRAIN_LOSS_CATEGORICAL_CROSSENTROPY. Using NNFW_TRAIN_LOSS_CATEGORICAL_CROSSENTROPY will ensure that the predicted input of loss is the result of performing softmax once regardless of whether the output of the model is the result of softmax or not.

Definition at line 193 of file onert-micro.h.

                          {.loss = NNFW_TRAIN_LOSS_CATEGORICAL_CROSSENTROPY,
                           .reduction_type = NNFW_TRAIN_LOSS_REDUCTION_SUM_OVER_BATCH_SIZE};

num_of_trainable_ops

int32_t nnfw_train_info::num_of_trainable_ops = NNFW_TRAIN_TRAINABLE_NONE

Number of layers to be trained from the back of the graph. Note that some values have special meaning. "-1" means that all layers will be trained. "0" means that no layer will be trained. Negative value less than -1 means error. The special values are collected in NNFW_TRAIN_NUM_OF_TRAINABLE_OPS_SPECIAL_VALUES enum.

Definition at line 257 of file nnfw_experimental.h.

num_trainble_ops

uint32_t nnfw_train_info::num_trainble_ops = 0

Definition at line 198 of file onert-micro.h.

opt

NNFW_TRAIN_OPTIMIZER nnfw_train_info::opt = NNFW_TRAIN_OPTIMIZER_ADAM

optimizer type

Definition at line 196 of file onert-micro.h.

Referenced by CfgRunner.CfgRunner::run().

---

The documentation for this struct was generated from the following files:

• onert-micro/onert-micro/include/onert-micro.h
• runtime/onert/api/nnfw/include/nnfw_experimental.h