Assignment 1
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
Assignment 1
Report Submission Due: 23 Jan, 2024 (before 11:59 pm)
Submission: Submit to NTULearn (Assignments à Assignment 1 à Report) with subject “ MH6812-Ass1-YourStudentID”.
Topic: Deep Learning models for Sentiment Classification
In this assignment, we will implement different deep learning models for sentiment classification task.
Code base for the Assignment 1: https://colab.research.google.com/drive/1- zXNgxFIGeTKuw2s2ftBscCKYCUnF20n?usp=sharing
In the code base, you are given a sample RNN model to classify the sentiment of sentences in iMDB dataset.
Tasks:
1. Warm up: Read, understand, and reimplement the example in the code base.
2. Conduct experiments with different optimizers: SGD, Adam, Adagrad and record the experimental results
3. Use Adam optimizer, conduct experiments with different number of epochs: 5, 10, 20, and 50.
4. Use Adam optimizer and 50 epochs, download and use pretrained Word2Vec embeddings as initialization of the models; compare the performance with the previous one.
Important notes:
- You can refer to here https://stackoverflow.com/questions/49710537/pytorch- gensim-how-to-load-pre-trained-word-embeddings on how to initialize the model using Word2Vec
- Note that you can pre-train Word2Cec on this dataset or use the pretrained Word2Vec available online. Available Pretrained Word2Vec can be downloaded
here:
https://drive.google.com/file/d/0B7XkCwpI5KDYNlNUTTlSS21pQmM/edit?usp=s haring
- Remember to match the vocabulary of our TEXT with Word2Vec vocab:
https://radimrehurek.com/gensim/models/word2vec.html
5. Use Adam optimizer, 50 epochs and randomly initialized embeddings, run the experiments with the following models:
• One-layer feed forward neural network, hidden dimension is 500.
• Two-layer feed forward neural network, hidden dimensions are 500 and 300.
• Three-layer feed forward neural network, hidden dimensions are 500, 300, and 200
• CNN model (using three feature maps with the sizes are 1, 2, and 3)
• LSTM model
• Bi-LSTM model
Report:
. Summarize the results of experiments (better in tables)
. Analysis, comparison, and explanation about the results (e.g., why there is difference between with and without Word2Vec? why this model is better than another model? i.e.)
. The format is free style. Try to be concise and not more than 4 pages
. The deadline is 11:59pm, 23 January 2023
2024-01-24