C19 Update NN World US Italy AR

Improvements today include (1) changing to Nadam optimizer, (2) plotting error on log scale to better judge overfitting (significant separation of training and validation curves), and (3) dropping number of units per hidden layer from 64 to 55 to avoid overfitting which was becoming evident on the Italy example. 

Model: "sequential"

_________________________________________

Layer (type)                 Output Shape              Param #   

=========================================

dense (Dense)                (None, 1)                 2         

_________________________________________

dense_1 (Dense)              (None, 55)                110       

_________________________________________

dense_2 (Dense)              (None, 55)                3080      

_________________________________________

dense_3 (Dense)              (None, 55)                3080      

_________________________________________

dense_4 (Dense)              (None, 55)                3080      

_________________________________________

dense_5 (Dense)              (None, 55)                3080      

_________________________________________

dense_6 (Dense)              (None, 55)                3080      

_________________________________________

dense_7 (Dense)              (None, 1)                 56       

=========================================

Total params: 15,568

Trainable params: 15,568

Non-trainable params: 0

Immature case: World

Semi mature case: US

Mature case: Italy

Accelerating case: Arkansas