LSTM and CNN Parameter Tuning results

Published: 6 September 2021| Version 2 | DOI: 10.17632/jtv3f9pbdb.2


The published data are part of the virtual-coach Table Tennis shadow-play training system developing results. Two deep models (LSTM and 2DCNN) were trained and tested based on the self-collected Table Tennis Forehand strokes sensory data The Models' parameters are tuned with the following values: 1) The Epoch number in range (100, 250, and 500), 2) The batch size in range (10, 50,100, 500 and1000), 3) The number of the LSTM and CNN layer in range (1, 2, and 3), 4) The number of filters in range (16 to 256), 5) The rate of dropout in range (0.1, 0.2, 0.3, ..., 0.9), 6) The number of the denes layer in range (1, 2, and 3), and 7) The number of the neurons in the dense layer is in the range (8 to 64) The published results contain all possible models' performance on the testing dataset.


Steps to reproduce

The LSTM and 2DCNN models are trained offline on a computer equipped with 3.7 GHz i7-8700k core processors, 32G RAM, and NVIDIA 1080ti GPU. To implement the LSTM and 2Dimentinal-CNN (2D-CNN), we applied the Tensorflow machine-learning library.


University of Tabriz


Machine Learning