Spam Text Messages

John Smart's old phone receives over a thousand text messages, and he needs to know which are spam. Each message is labelled spam or legit.

Task 1 (8 points): count the unique words (tokens), count spam and legit messages, clean the texts with the provided clean_text function (unchanged) and add the cleaned texts as a new DataFrame column. Task 2 (10 points): write a tokenizer assigning a unique index to every word, convert every message to an index sequence, pad all sequences to the length of the longest message, encode labels as spam = 0, legit = 1, and add the sequences and labels to the DataFrame. Task 3 (3 points): implement a PyTorch Dataset for the sequences and labels.

Task 4 (5 points): build the network shown in the figure (vocabulary size 9476 + PAD = 9477, EMBEDDING_DIM = 16, two output classes) (4) and load the weights from jános_email_modell.pth (1). Task 5 (4 points): classify a given suspicious message about a Himalayan hiking trip (2) and write a malicious message that the model classifies as legit (2).