Read Me

Using neural networks to solve No Thanks!

Exercise 01 (2024-02-18)

No Thanks parameters

• Cards: 1-5, omit 1
• Chips: 1

NN parameters

• Target: value
• Structure: 24 x 100 -> 100 x 100 -> 100 x 1
• Learning rate: 5e-3
• Batch size: 256
• Loss fn: MSELoss
• Epochs: 5000
• Train on all data -- no test
• Train loss: ~0.15

Exercise 02 (2024-02-18)

No Thanks parameters

• Cards: 1-5, omit 1
• Chips: 1

NN parameters

• Target: value

Exercise 03 (2024-02-18)

Exercise 04 (2024-02-19)

• Attempt to overfit a single batch
• Done
• Can get very close to perfect.
• Train loss: ~0.01

Exercise 03

• Compare train loss and test loss
• Split (action) data into 75% train, 25% test

NN parameters

Target: action Train/test: 75%/25% Learning rate: 5e-3 Epochs: 6000 Batch size: 1000 Hiden: 100

Saved model

2024-02-19_no_thanks_nn_03.pth

Exervise 06 - Decision Tree Model