CrfLossLayer.md

CCrfLossLayer Class

• CCrfLossLayer Class
  • Settings
    • Loss weight
    • Gradient clipping
  • Trainable parameters
  • Inputs
  • Outputs
    • Getting the value of the loss function

This class implements a layer that calculates the loss function used for training a CRF. The value of the loss function is equal to -log(probability of the correct class sequence)

Settings

Loss weight

void SetLossWeight( float lossWeight );

Sets the multiplier for this function gradient during training. The default value is 1. You may wish to change the default if you are using several loss functions in your network.

Gradient clipping

void SetMaxGradientValue( float maxValue );

Sets the upper limit for the absolute value of the function gradient. Whenever the gradient exceeds this limit its absolute value will be reduced to GetMaxGradientValue().

Trainable parameters

The layer has no trainable parameters.

Inputs

The layer has 3 inputs.

The first input accepts a blob with int data that contains optimal class sequences. The dimensions of the blob are:

• BatchLength is equal to the network inputs' BatchLength
• BatchWidth is equal to the network inputs' BatchWidth
• Channels is the number of classes
• the other dimensions are equal to 1

The second input accepts a blob with float data containing non-normalized logarithm of probabilities of the optimal class sequences. The blob has the same dimensions as the first input.

The third input accepts a blob with non-normalized logarithm of probability of the correct class being in this position. The blob dimensions are:

• BatchLength is equal to the network inputs' BatchLength
• BatchWidth is equal to the network inputs' BatchWidth
• the other dimensions are equal to 1

Outputs

This layer has no output.

Getting the value of the loss function

float GetLastLoss() const;

Use this method to get the value of the loss function calculated on the network's last run.