Reference no: EM133868350
Deep Learning
Assignment Report
Task 1
Task 1a
Data issues
...
Task 1b
Why not use random_split?
...
Task 1c
Reduce epoch time
...
Confusion matrix
...
Task 1d
Account for data issues
...
Task 1e
Vertical Flips
...
Effect of Augmentation
...
[Challenge] 5 crop augmentation
...
Task 1f
Experiments
<See attached excel document>
<See attached Weights and Bias Report>
Write a discussion about the key findings from the experimental results.
...
[Challenge] Batch size
...
Task 2
Task 2a
Data issues
...
Task 2b
Similar embeddings
...
Task 2c
Saved model weights
...
What do the longs (int64) represent?
...
[Challenge] Visualization of fine-tuned DistillBERT model.
...
Task 2d
Class distributions and learning rate
...
Relative performances before and after fixing learning rate
This is the files you will need for completing assignment 2. These files are:
1. Assignment.ipynb - This is the file that you need to fill in codes and run them to obtain results.
2. datasets.py / explore.py / models.py / train.py / sc1.py / visualize_embeddings.py - These are the python files the you also need to complete in the code. If you are using Google Colab, please upload them to the Colab after filling in your own code, so that your code in the ipynb file can import them. One way to upload is: in the Colab, click on the file symbol on the left, and then upload these python files into the popped out directory. Please note that these files will be deleted if you close the Colab. You will need to upload again next time you open the .ipynb on Colab. Get in touch with us for online assignment help service!
3. CSE5DL Assignment Report - This is the template for your report.
This assignment has 2 major tasks, and they can be challenging but not impossible. You will have clear idea of completing these tasks if you practice well on the lab.
For submission, you can upload necessary files by packaging them into a zip file. Here are the things you need to submit:
1. Code in the ipynb notebook filled out.
2. Fill in the TODO code in the following python files:
train.py
models.py
datasets.py
visualize_embeddings.py
3. Written report
4. Wandb report
5. Excel spreadsheet with 10 training runs for Task 1, including the following information:
Hyper parameters
Model configuration
Results
For marking, I will mainly look at your .ipynb file and your report but will also look into the other susbmitted files carefully if necessary. Your .ipynb file is supposed to contain the running results of your codes without bugs.
For questions or confusions on the tasks, feel free to ask me on lab or via emails (but be aware that it is not appropriate to directly ask how to complete the tasks, since this will be unfair to others if I tell you the potential solutions).