Click or drag to resize

GraphXlaSparseDenseMatmulGradWithFtrlAndCsrInput Method

https://www.emgu.com/wiki/index.php/Emgu_TF
XlaSparseDenseMatmulGradWithFtrlAndCsrInput

Namespace:  Emgu.TF
Assembly:  Emgu.TF (in Emgu.TF.dll) Version: 2.16.1.1620+d2904aaafaa66327b642f4fbd5e3c7133dd331e0
Syntax
public Operation XlaSparseDenseMatmulGradWithFtrlAndCsrInput(
	Output row_pointers,
	Output sorted_sample_ids,
	Output sorted_token_ids,
	Output sorted_gains,
	Output activation_gradients,
	Output learning_rate,
	Output embedding_table,
	Output accumulator,
	Output linear,
	Output num_minibatches_per_physical_sparse_core,
	bool multiply_linear_by_learning_rate,
	float beta,
	float learning_rate_power,
	float l1_regularization_strength,
	float l2_regularization_strength,
	string table_name,
	float clip_weight_min = -∞f,
	float clip_weight_max = ∞f,
	string opName = "XlaSparseDenseMatmulGradWithFtrlAndCsrInput"
)

Parameters

row_pointers
Type: Emgu.TFOutput
Input to the operation.
sorted_sample_ids
Type: Emgu.TFOutput
Input to the operation.
sorted_token_ids
Type: Emgu.TFOutput
Input to the operation.
sorted_gains
Type: Emgu.TFOutput
Input to the operation.
activation_gradients
Type: Emgu.TFOutput
Input to the operation.
learning_rate
Type: Emgu.TFOutput
Input to the operation.
embedding_table
Type: Emgu.TFOutput
Input to the operation.
accumulator
Type: Emgu.TFOutput
Input to the operation.
linear
Type: Emgu.TFOutput
Input to the operation.
num_minibatches_per_physical_sparse_core
Type: Emgu.TFOutput
Input to the operation.
multiply_linear_by_learning_rate
Type: SystemBoolean
multiply linear by learning rate
beta
Type: SystemSingle
beta
learning_rate_power
Type: SystemSingle
learning rate power
l1_regularization_strength
Type: SystemSingle
l1 regularization strength
l2_regularization_strength
Type: SystemSingle
l2 regularization strength
table_name
Type: SystemString
table name
clip_weight_min (Optional)
Type: SystemSingle
clip weight min
clip_weight_max (Optional)
Type: SystemSingle
clip weight max
opName (Optional)
Type: SystemString
The name of the operation

Return Value

Type: Operation
The operation, where: [0] updated_embedding_table(type: DtFloat). [1] updated_accumulator(type: DtFloat). [2] updated_linear(type: DtFloat).
See Also