Long-Term Memory Recallmedium

Long-Term Memory Recall

medium Build an Agent 15–20 min

Standard library only · runs in your browser

Trimming (stage 7) throws old turns away. Long-term memory keeps the important facts outside the context window and brings back only the ones relevant to the current question, newest first when facts conflict.

Task

Implement recall(memories, query, k=3). Each memory is {"id": str, "turn": int, "text": str}.

Terms of a text: lowercase it, take re.findall(r"[a-z0-9]+", ...), and drop single-character words (the "s" in "user's") and words in the provided STOPWORDS set. Use a set, so repeats count once.
Score of a memory: the number of query terms that also appear in the memory's terms.
Leave out memories that score 0.
Sort by score (high first), then turn (newer first), then id (alphabetical).
Return the ids of the top k.

Function Signature

recall(memories: list[dict], query: str, k: int = 3) -> list[str]

Examples

Example 1: Newer fact outranks the older one
Input: memories = [{'id': 'm1', 'turn': 1, 'text': 'User\'s name is Priya and she lives in Lima.'}, {'id': 'm2', 'turn': 3, 'text': 'Priya prefers vegetarian restaurants.'}, {'id': 'm3', 'turn': 4, 'text': 'The user\'s flight to Tokyo departs on March 3.'}, {'id': 'm4', 'turn': 6, 'text': 'Priya moved from Lima to Paris last month.'}, {'id': 'm5', 'turn': 7, 'text': 'Weather in Paris was rainy during the last trip.'}]
query = 'Which city is Priya in now: Lima or Paris?'
Output: ['m4', 'm1', 'm5']

Example 2: One clear match
Input: memories = [{'id': 'm1', 'turn': 1, 'text': 'User\'s name is Priya and she lives in Lima.'}, {'id': 'm2', 'turn': 3, 'text': 'Priya prefers vegetarian restaurants.'}, {'id': 'm3', 'turn': 4, 'text': 'The user\'s flight to Tokyo departs on March 3.'}, {'id': 'm4', 'turn': 6, 'text': 'Priya moved from Lima to Paris last month.'}, {'id': 'm5', 'turn': 7, 'text': 'Weather in Paris was rainy during the last trip.'}]
query = 'When does the Tokyo flight leave?'
Output: ['m3']

Example 3: No overlap recalls nothing
Input: memories = [{'id': 'm1', 'turn': 1, 'text': 'User\'s name is Priya and she lives in Lima.'}, {'id': 'm2', 'turn': 3, 'text': 'Priya prefers vegetarian restaurants.'}, {'id': 'm3', 'turn': 4, 'text': 'The user\'s flight to Tokyo departs on March 3.'}, {'id': 'm4', 'turn': 6, 'text': 'Priya moved from Lima to Paris last month.'}, {'id': 'm5', 'turn': 7, 'text': 'Weather in Paris was rainy during the last trip.'}]
query = 'What\'s the GDP of Chile?'
Output: []

Example 4: k limits the result
Input: memories = [{'id': 'm1', 'turn': 1, 'text': 'User\'s name is Priya and she lives in Lima.'}, {'id': 'm2', 'turn': 3, 'text': 'Priya prefers vegetarian restaurants.'}, {'id': 'm3', 'turn': 4, 'text': 'The user\'s flight to Tokyo departs on March 3.'}, {'id': 'm4', 'turn': 6, 'text': 'Priya moved from Lima to Paris last month.'}, {'id': 'm5', 'turn': 7, 'text': 'Weather in Paris was rainy during the last trip.'}]
query = 'Priya Paris'
k = 1
Output: ['m4']

Constraints

`STOPWORDS` is provided in the starter. Use it as-is.
Return a list of ids, not the memory dicts.
A query made only of stopwords recalls nothing.

Python3Saved ✓

import json

import math

import re

STOPWORDS = {"a", "an", "and", "are", "did", "do", "does", "for", "i", "in", "is",

"it", "me", "my", "of", "on", "the", "to", "was", "what", "when",

"where", "which", "who", "with", "you", "your"}

def recall(memories, query, k=3):

"""Return the ids of the k most relevant memories (score, then recency, then id)."""

# Your code here

pass

Case 1: Newer fact outranks the older one

Input: memories = [{'id': 'm1', 'turn': 1, 'text': 'User\'s name is Priya and she lives in Lima.'}, {'id': 'm2', 'turn': 3, 'text': 'Priya prefers vegetarian restaurants.'}, {'id': 'm3', 'turn': 4, 'text': 'The user\'s flight to Tokyo departs on March 3.'}, {'id': 'm4', 'turn': 6, 'text': 'Priya moved from Lima to Paris last month.'}, {'id': 'm5', 'turn': 7, 'text': 'Weather in Paris was rainy during the last trip.'}]
query = 'Which city is Priya in now: Lima or Paris?'

Expected: ['m4', 'm1', 'm5']

Case 2: One clear match

Input: memories = [{'id': 'm1', 'turn': 1, 'text': 'User\'s name is Priya and she lives in Lima.'}, {'id': 'm2', 'turn': 3, 'text': 'Priya prefers vegetarian restaurants.'}, {'id': 'm3', 'turn': 4, 'text': 'The user\'s flight to Tokyo departs on March 3.'}, {'id': 'm4', 'turn': 6, 'text': 'Priya moved from Lima to Paris last month.'}, {'id': 'm5', 'turn': 7, 'text': 'Weather in Paris was rainy during the last trip.'}]
query = 'When does the Tokyo flight leave?'

Expected: ['m3']

Case 3: No overlap recalls nothing

Input: memories = [{'id': 'm1', 'turn': 1, 'text': 'User\'s name is Priya and she lives in Lima.'}, {'id': 'm2', 'turn': 3, 'text': 'Priya prefers vegetarian restaurants.'}, {'id': 'm3', 'turn': 4, 'text': 'The user\'s flight to Tokyo departs on March 3.'}, {'id': 'm4', 'turn': 6, 'text': 'Priya moved from Lima to Paris last month.'}, {'id': 'm5', 'turn': 7, 'text': 'Weather in Paris was rainy during the last trip.'}]
query = 'What\'s the GDP of Chile?'

Expected: []

Case 4: k limits the result

Input: memories = [{'id': 'm1', 'turn': 1, 'text': 'User\'s name is Priya and she lives in Lima.'}, {'id': 'm2', 'turn': 3, 'text': 'Priya prefers vegetarian restaurants.'}, {'id': 'm3', 'turn': 4, 'text': 'The user\'s flight to Tokyo departs on March 3.'}, {'id': 'm4', 'turn': 6, 'text': 'Priya moved from Lima to Paris last month.'}, {'id': 'm5', 'turn': 7, 'text': 'Weather in Paris was rainy during the last trip.'}]
query = 'Priya Paris'
k = 1

Expected: ['m4']