Detect a Repeating Action and Break the Loop
The problem, from first principles
A model stuck in a bad pattern doesn't always fail loudly — sometimes it just calls the exact same tool with the exact same arguments over and over, each time apparently confident this attempt will be different, never actually making progress. This looks completely different from the "consecutive failures" pattern in the earlier self-reflection question: the observations here might not look like failures at all (the tool might succeed every time!) — the problem is purely that the same action keeps recurring, which is its own distinct signal that something's wrong.
From theory to code
You're given every step's (action, action_input) pair, in order, a window size, and a repeat_threshold. Implement detect_repeating_action(actions, window, repeat_threshold). Walk the steps in order. After each step, look at only the last window steps seen so far (fewer if not enough steps have happened yet) and count how many of them exactly match the action just taken. The moment that count reaches repeat_threshold, return the current 1-indexed step number. Return None if it never happens.
Constraints
- 1 to 500 steps;
window >= 1,repeat_threshold >= 1.
Hints
Hint 1
A fixed-size sliding window is exactly what collections.deque(maxlen=window) gives you for free — appending past capacity automatically drops the oldest entry, so you never need to manually trim anything.
Hint 2
Count matches within the current window only, not across the whole history — a deque with maxlen set already only ever contains the most recent window items, so counting within it is automatically counting within the window.
Keywords
No results yet
Run your code to check it against the examples, or submit to run every test.