Build a Python Survey Response Summarizer

Select a timestamp to watch that moment on YouTube.

1. 00:00

   We have three survey rows. One respondent chose Python and gave a rating, one skipped the choice entirely, and one typed the same choice with extra spaces and different capitalization. How many votes should the summary show? Before we calculate anything, decide what counts as an answer. The project runs on a local list of dictionaries, with no external service or private survey upload. That makes it safe to experiment with missing values and invalid data. Our target output has two fields: a mapping of normalized choice counts and an average rating when at least one valid rating exists. Those fields answer different questions and should be computed from different

2. 00:41

   subsets of the input. A survey export is not guaranteed to contain a response for every question. Someone may leave choice blank, type extra spaces, or use different capitalization. A useful summary counts actual choices after normalization and reports the average of valid numerical ratings. This project uses a list of dictionaries in memory, so it works in the browser without uploading anyone's answers. We deliberately keep counts and average separate: a missing choice should not invent a vote, and a missing rating should not enter the denominator. The broken implementation indexes the choice key in every row.

3. 01:24

   It works for the first respondent, then stops at the empty dictionary with KeyError. The traceback points to the indexing expression, not to Counter itself. It would be easy to add a dummy choice value to every test row and make the error disappear, but that would hide the real input shape. Survey questions are often optional. The code must represent that policy intentionally. Even without the missing row, counting the raw strings would split Python and a spaced lower-case version into separate labels. The broken run therefore exposes two distinct problems: assuming a key always exists, and treating formatting differences as different responses.

4. 02:06

   The broken program indexes row['choice'] for every record. When a respondent skips that question, it raises KeyError and produces no summary at all. Even if every record had a key, 'Python', ' python ', and 'PYTHON' would be counted as different choices without normalization. Diagnose the raw keys, values, and types before tallying. Missing data is not the same as the string 'None' or a numerical zero. A summary must state what it omits rather than silently converting absent values into a real answer. The diagnosis walks the records and prints each row's keys, the raw choice, and its stripped, case-folded value.

5. 02:48

   This is more informative than adding a broad exception handler around the entire report. For the missing row, get returns an empty string. For the spaced response, strip removes boundary whitespace and casefold normalizes capitalization. We can now state a rule: count a choice only if meaningful text remains after normalization. Ratings need their own rule. A missing or null choice should be skipped. A missing rating should be absent from the average, while a supplied value outside the agreed one-to-five integer range should raise an error. Silent coercion would make it hard to spot a bad export. Use Counter for normalized nonempty choices.

6. 03:30

   Strip surrounding whitespace, apply casefold to make case-insensitive groups, and increment only when text remains. For ratings, accept integers one through five while explicitly rejecting bool, because bool is a subclass of int in Python. Store valid ratings in a separate list and use statistics.mean when that list is nonempty. Return None when no ratings were supplied. Sort the count mapping for stable display and tests. Validation errors should identify malformed supplied ratings rather than dropping them without notice. The fixed function uses Counter for choices and a separate list for ratings.

7. 04:10

   Each nonempty normalized choice increments its count. Each supplied rating is checked as an actual integer between one and five. We use type(value) is int because Python booleans behave like integers in some contexts, yet True is not a legitimate survey rating. The mean is computed only when at least one rating was observed; otherwise the result is None. The count mapping is sorted before returning it, giving a stable presentation order that is useful for the webpage and tests. The sample now combines both Python spellings into a count of two and averages the two real ratings. Read the loop as two independent pipelines.

8. 04:52

   The choice pipeline obtains optional text, normalizes it, and counts only a nonempty result. The rating pipeline checks whether a value was supplied, verifies its exact type and allowed range, and adds it to a separate list. At return time, dict(sorted(counts.items())) makes output predictable, while mean(ratings) if ratings else None preserves the difference between an unobserved average and a numerical value. Notice that we never divide by the total number of survey rows. That number includes respondents who did not rate, so using it as the denominator would bias the result. This function is small enough to audit yet captures several common data-cleaning

9. 05:34

   decisions. The verification run tries a blank-only survey and a boolean rating. Blank and missing choices produce no vote, and the absence of ratings produces None instead of a fabricated zero. A boolean rating triggers ValueError, which makes the malformed supplied value visible to the caller. Try adding a fourth response with a new choice and no rating: the count changes, but the average should not. If this were a real product, we would also explain its normalization and missing-data policy next to every chart. On the lesson page, complete the function, run the practical checker, and then use

10. 06:16

    the quiz to explain why missing, blank, zero, and invalid are not interchangeable. The verification run includes an entirely blank survey, an explicit null choice, and a record with no rating. It should report an empty count mapping and no average, not divide by zero or show a fabricated zero-star result. For real survey data, consider privacy, sample selection, and whether response categories should be restricted to a known list. This tutorial focuses on the data-handling mechanics. The task checker tests normalization, missing answers, and an invalid boolean rating, so the same code can be trusted for those specified cases rather than just

11. 06:56

    one happy-path sample.