Scribd
Upload
15 views

NLP Assignment 2

Nlp assignment

Uploaded by

Arkish Chakraborty
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
15 views

NLP Assignment 2

Nlp assignment

Uploaded by

Arkish Chakraborty
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Course instructor: Dr.

Jasabanta Patro Assignment number: 2


Course: DSE 407/607: NLP Date: October 7, 2024
Marks: 10 Date of submission: October 15, 2024

Regulations:

• Each student is required to submit solutions based on the specified task.

• Multiple submissions are not allowed.

• Plagiarism: Strictly prohibited. All work should be original. The code and report will
be checked for plague (as well as AI detector) and appropriate action will taken if found
guilty of copying.

Submission Guidelines:

• Deliverables: public URL of (i) code (with proper comments).

– The Colab notebook should only contain the inference part of the model and load
the pre-trained weights. The training part should be commented out.
– The model should be able to load weights from your public GitHub repository (create
a repo with the trained model and download weights from it).

• File naming convention:


rollno name nlpassignment1.ipynb
rollno name nlpassignment1.pdf

• Students need to submit only the URL of the Colab notebook (with public access) with
clear instructions for running the code. The runtime of the code should not be more than
10 minutes.

• Deadline: All assignments must be submitted by the deadline. Late submissions will be
penalized.

Marking:

• Marking will be done based on two criterias, (i) code, and (ii) model performance
(more focus will be given to performance).

• The performance of each submission will be evaluated using average macro F1-score based
on the predicted labels and the gold ones.
2

• All submitted code should be reproducible with public access. If the results cannot be
reproduced, the submission will be considered incomplete and the submission will not be
marked.

Text Classification Using FFNN, RNN and LSTM:


Given a target text snippet, predict the perceived emotion(s) of the speaker. Specifically, select
whether each of the following emotions apply: joy, sadness, fear, anger, surprise, or disgust. In
other words, label the text snippet with: joy (1) or no joy (0), sadness (1) or no sadness (0),
fear (1) or no fear (0), anger (1) or no anger (0), surprise (1) or no surprise (0).
Further details on the task (Track-A) is provided in the following URL:

https://github.com/emotion-analysis-project/SemEval2025-task11

The sample dataset is provided in the above URL. Students can refer it to understand the
task and the dataset format. The train dataset is available on the URL : https://github.
com/debajyotimaz/nlp_assignment. Students are supposed to submit the code of the best
performing algorithm and that algorithm will be evaluated on our custom test set (which will
be kept private). Use this colab notebook to get started: https://colab.research.google.
com/drive/13yNxUnB866IqHF8H1mnXIRQYT-t4COBA?usp=sharing.

Task overview:

1. Embeddings: Create a custom Word2Vec model from scratch. Train it on your dataset
with an embedding size of 100 dimensions, and save the trained embeddings in your
GitHub repository.

2. Modeling: Feedforward Neural Networks (FFNN), Recurrent Neural Networks (RNN),


and Long Short-Term Memory networks (LSTM) are permitted, and combinations of
these architectures can also be applied.

• Use FFNN with a maximum size of 64 units.


• Use RNN or LSTM layers with a maximum size of 64 units and a sequence length
of up to 128 tokens.

3. Training setup:

• Use Adam, AdamW or SGD optimizer.

4. Note: Comment out the training part (both embedding creation and model training)
of the code (we can undo comments to check the training part also). The model should
already be trained, and the deliverable will be focused on inference only.

2
3

Constraints Value
Maximum number of layers 4
Maximum number of units per layer 64
Maximum embedding size 100
Maximum sequence length 128
Optimizer Adam, AdamW, SGD
Learning rate 0.001

Table 1: Modeling constraints

You might also like