C19 Update with Brazil + World NN

2020-06-22

Total US C19 deaths 119,969

Last five daily = [754 715 692 595 250]

Future estimates

   124143 +/- 1587  =  Jul 1 total (95% confidence)

   133408 +/- 1587  =  Aug 1

   137881 +/- 1587  =  Sep 1

Total world C19 deaths 468,331

Last five daily = [5274 5020 6038 4448 3866]

Future estimates

   486789 +/- 176676  =  Jul 1 total (95% confidence)

   571269 +/- 176676  =  Aug 1

   636381 +/- 176676  =  Sep 1

Total Brazil C19 deaths 50,591

Last five daily = [1269 1238 1206 1022  615]

Future estimates

   59184 +/- 10936  =  Jul 1 total (95% confidence)

   80809 +/- 10936  =  Aug 1

   91339 +/- 10936  =  Sep 1

Total Arkansas C19 deaths 225

Last five daily = [ 9 11  6 10  1]

Further experimentation with NN code leads to best result with one hidden layer of 15 neurons, random_state=5, and the model is saved for application to future data. Using a random state ensures the result is reproducible, as I have confirmed on several runs using the data as of today. When a new data point comes in tomorrow I expect the fit to update when the saved NN model is applied. 

The choice random_state=5 is near the best score of 0.92 while maintaining a smooth function. Subjective, of course.

Score =  0.9195

Neural Net: MLPRegressor(activation='tanh', alpha=0.0001, batch_size='auto', beta_1=0.9,

             beta_2=0.999, early_stopping=False, epsilon=1e-08,

             hidden_layer_sizes=(15,), learning_rate='constant',

             learning_rate_init=0.001, max_fun=15000, max_iter=2000,

             momentum=0.9, n_iter_no_change=10, nesterovs_momentum=True,

             power_t=0.5, random_state=5, shuffle=True, solver='lbfgs',

             tol=0.0001, validation_fraction=0.1, verbose=False,

             warm_start=False)

2020-06-22

Total World C19 deaths 468,322

Last five daily = [5274 5020 6038 4448 3874]

Future estimates

   5441  =  Jul 1 daily (day 161)

   6054  =  Aug 1 daily (day 192)

   6110  =  Sep 1 daily (day 223)