At the referee’s request, we performed a test how our neural network architecture would perform with randomized phase. For this test we created a completely new training and testing dataset where the phase of the maximum/minimum value in the 0–1 interval was randomized. In this case the light curve minima of the eclipsing binaries and the maxima of the pulsating variable stars should not have any external information. Our results show that our neural network does not learn the phase information, only the light curve shape.