Week 23 - Modifying training samples number

Modify training samples number analysis

We propose a new experiment in which we analyze the impact of the number of samples used in training on the predictive capacity of the network. For this I have generated training datasets, which follow the 1 DOF linear motion, with different number of samples in each one and I have trained the CNN structure. To test the networks I have used a fixed set of 1000 samples in such a way that the comparison is fair.

In the graph it can be seen how increasing the number of training samples produces a positive effect on the network performance up to a limit, from which performance is maintained. This limit will depend on the complexity of the motion.