# 機器學習 2021 HW_1 note
colab:
https://colab.research.google.com/drive/1ME3xTuo3GswWlrEBfQDen7tAUyNAq6Hf#scrollTo=aQikz3IPiyPf

TODO:
1. How to modify this model to achieve better performance?
2. you may implement L1/L2 regularization here
3. Using 40 states & 2 tested_positive features
4. How to tune these hyper-parameters to improve your model's performance?

training time

### 筆記
dataset:
train train 2700
dev validation
test test 893

### 想做的事

把所有可變得參數弄成儀表板,一個一個調整
和數組輸入功能,直接輸入整個數組

### record:

| title | descripe | training lost | stop point (epoch) | training time |
| - | - | - | - | - |
| sample code | sample code | 0.7371 | 1437 | 33.4 |
| sample code | sample code | 0.7763 | 1441 | 31.24 |

Saving model (epoch = 106, loss = 0.7763)
Finished training after 307 epochs
training time: 6.7936179637908936second

no Normalize
Saving model (epoch = 29, loss = 58.9434)
Finished training after 230 epochs
training time: 5.248743057250977second

Saving model (epoch = 1323, loss = 0.7598)
Finished training after 1524 epochs
training time: 34.106801986694336second

target only
Saving model (epoch = 318, loss = 0.9614)
Finished training after 519 epochs
training time: 10.781646013259888second