casval5fn28/nycu-2022-introduction-to-machine-learning-final-project

NYCU 2022 Introduction to Machine Learning Final Project

Environment

• python: 3.10.5
• tesorflow: 2.11.0
• keras: 2.11.0

Process of reproducing the work

• First install following packages(for google colab)

!pip install keras
!pip install tensorflow

• Input data is in the same directory as two ipynb files , you can move them to desired places and specify the path in the two ipynb files just mentioned , the place to adjust is commented

• To reproduce 109550135.csv(file for submission) without retraining :

1. Download Pretrained Model
2. Put the downloaded model file in the same directory as 109550135_Final_inference.ipynb
3. Run 109550135_Final_inference.ipynb

• To reproduce 109550135.csv(file for submission) from the beginning :

1. Run 109550135_Final_train.ipynb and get 109550135_model.h5
2. 109550135_model.h5 will be in the same directory as 109550135_Final_train.ipynb and 109550135_Final_inference.ipynb
3. So you can run 109550135_Final_inference.ipynb directly
4. A 109550135.csv will be created in the same directory as two ipynb files , it's the result for submission

Train the model

Once you build up the environment, you can run 109550135_Final_train.ipynb for training the model.

Of course , you can skip the former step and use 109550135_Final_inference.ipynb directly to evaluate performace of the model

P.S. Don't forget to download Pretrained Model to the same directory as 109550135_Final_train.ipynb first, then you can run 109550135_Final_inference.ipynb.

Evaluate model performance

Evaluate model performace by 109550135_Final_inference.ipynb.

If you want to evaluate the model performance without retraining , you need to download Pretrained Model to the same directory as 109550135_Final_train.ipynb first, then run 109550135_Final_inference.ipynb.

Prediction result

After finishing evaluation of the model by 109550135_Final_inference.ipynb, you will get 109550135.csv , which is the result you need .

The below is the accuracy result:

Submission	Private Score	Public Score
109550135.csv	0.59125	0.59547