Research on the Evaluation Model of Dance Movement Recognition and Automatic Generation Based on Long Short-Term Memory

Table 7

Experimental environment settings.

All the experiments were completed on a NVIDIA 1080tigpu
Generator training stage	Set the training model to train 10000 rounds. The model input dimension is 35, the encoder convolution layers are 3, the maximum length of each convolution kernel is 5, the decoder RNN dimension is 1024, the prenet dimension is 256, the learning rate is set to 0.001, the gradient clipping threshold is set to 1, the weight attenuation is set to 1e-6, the batchsize is set to 40, the seqlen is set to 125, and the optimizer uses Adam.
Discriminator training stage	The discriminator is trained once every three training rounds, the learning rate of the discriminator is 0.001, the weight attenuation is 1e-6, and the optimizer uses Adam.
The training set used 80% dance data, and the test set used 20% dance data