C19 Update with NN Improvement

 

Modified my neural network codes for US and world modeling of total C19 deaths. Problem was that I often had to rerun the code a few times to get an accurate result. Fix was to set up an accuracy target (I chose 2000 for world, 1000 for US) and a loop that checks the accuracy after model training and evaluation on the entire data set. If the acc is above my target, clear the NN model, train, evaluate, check acc against target — and repeat till acc is below the target value. Early tests show it getting there in less than 10 iterations. Nice.

As always, remember the NN fit is done on the total deaths curve and the daily curve is computed from that.