# BEGIN - DO NOT CHANGE THESE IMPORTS/CONSTANTS OR IMPORT ADDITIONAL PACKAGES.

from torch.utils.data import Dataset

# END - DO NOT CHANGE THESE IMPORTS/CONSTANTS OR IMPORT ADDITIONAL PACKAGES.

# HeadlineDataset

# This class takes a Pandas DataFrame and wraps in a Torch Dataset.

# Read more about Torch Datasets here:

# https://pytorch.org/tutorials/beginner/basics/data_tutorial.html

#

class HeadlineDataset(Dataset):

# initialize this class with appropriate instance variables

def __init__(self, vocab, df, max_length=50):

# For this method: We would *strongly* recommend storing the dataframe

# itself as an instance variable, and keeping this method

# very simple. Leave processing to __getitem__.

#

# Sometimes, however, it does make sense to preprocess in

# __init__. If you are curious as to why, read the aside at the

# bottom of this cell.

#

## YOUR CODE STARTS HERE (~3 lines of code) ##

return

## YOUR CODE ENDS HERE ##

# return the length of the dataframe instance variable

def __len__(self):

df_len = None

## YOUR CODE STARTS HERE (1 line of code) ##

## YOUR CODE ENDS HERE ##

return df_len

# __getitem__

#

# Converts a dataframe row (row["tokenized"]) to an encoded torch LongTensor,

# using our vocab map created using generate_vocab_map. Restricts the encoded

# headline length to max_length.

#

# The purpose of this method is to convert the row - a list of words - into

# a corresponding list of numbers.

#

# i.e. using a map of {"hi": 2, "hello": 3, "UNK": 0}

# this list ["hi", "hello", "NOT_IN_DICT"] will turn into [2, 3, 0]

#

# returns:

# tokenized_word_tensor - torch.LongTensor

# A 1D tensor of type Long, that has each

# token in the dataframe mapped to a number.

# These numbers are retrieved from the vocab_map

# we created in generate_vocab_map.

#

# **IMPORTANT**: if we filtered out the word

# because it's infrequent (and it doesn't exist

# in the vocab) we need to replace it w/ the UNK

# token

#

# curr_label - int

# Binary 0/1 label retrieved from the DataFrame.

#

def __getitem__(self, index: int):

tokenized_word_tensor = None

curr_label = None

## YOUR CODE STARTS HERE (~3-7 lines of code) ##

## YOUR CODE ENDS HERE ##

return tokenized_word_tensor, curr_label

#

# Completely optional aside on preprocessing in __init__.

#

# Sometimes the compute bottleneck actually ends up being in __getitem__.

# In this case, you'd loop over your dataset in __init__, passing data

# to __getitem__ and storing it in another instance variable. Then,

# you can simply return the preprocessed data in __getitem__ instead of

# doing the preprocessing.

#

# There is a tradeoff though: can you think of one?

#