NCA-GENL Data Analysis and Visualization Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  A pie chart of documents bucketed into 1K-token bins.

  Why it's wrong here

  A pie chart with eight or more slices is hard to read and cannot show cumulative truncation impact. It shows only the proportion in each bin, not how many documents would survive a given cutoff, so it does not directly support choosing max_sequence_length. Pie charts also obscure the long tail, which is the very region of interest here.

• ✓

  A cumulative distribution function (CDF) plot of token lengths with a vertical marker at each candidate max_sequence_length.

  Why this is correct

  A CDF directly answers 'what fraction of documents are at or below X tokens', which is exactly the decision needed for max_sequence_length. Marking 2,048, 4,096, and 8,192 on the CDF lets the data scientist read off the percentage of the corpus that would be truncated, making the trade-off between memory footprint and data loss explicit.

• ✗

  A word cloud of the most frequent tokens in the longest documents.

  Why it's wrong here

  A word cloud visualizes lexical frequency, not length distribution, so it provides no information about how many documents exceed a candidate truncation threshold. It also discards positional and length information entirely. Using it here would answer a content question instead of the sizing question the pipeline actually needs.

• ✗

  A box plot of token lengths grouped by document source.

  Why it's wrong here

  A box plot summarizes median, quartiles, and outliers, but with a heavy right skew the whiskers and quartiles hide the exact tail mass beyond 8,192 tokens. It answers 'how variable are sources' rather than 'what fraction of documents fall below a candidate cutoff', so it cannot directly justify a max_sequence_length value.