Behind the scenes the test is proceeding. In this link you can find continuous updates
https://vafed.it/benchmark_tp_1.php. I am getting 3% (Expectancy) from 24 December 2019 to today 8 April 2020. Meanwhile the tests continue. This weekend there will be an update to the training of the model reaching 200000 test images. He is now working with 150000 training images. There is one update every approximately 50000 images. I keep you updated.
I enjoy projects like this. But I don't understand why you spend training on pre-existing images? If you were working with the raw data you wouldn't have to modify your algorithm for each chart that didn't use the same scales. After you spend all that time processing images, you still come down to change per period, which you already have with the raw data. :/