docs: Fix the precision calculation in context-precision metric doc #685
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
shahules786
approved these changes
Feb 29, 2024
joy13975
pushed a commit
to joy13975/ragas
that referenced
this pull request
Mar 4, 2024
…xplodinggradients#685) Precision@1 = 0/1 != 1. It should be 0. This PR fixes the calculation of Precision@1 and subsequently `\text{Context Precision} = {\text{(0+0.5)} \over \text{2}} = 0.25`  Co-authored-by: Amit Timalsina <amit@kniru.com>
shahules786
pushed a commit
that referenced
this pull request
Mar 4, 2024
Eventhough @amit-timalsina made a fix to the Context Precision docs a few day ago in #685 the description is still wrong in multiple ways. 1) The total number of **relevant items** is used, not all items. That is, in the example calculation the denominator needs to be 1 not 2, leading to a result of 0.5 not 0.25. 2) The formula for Context Precision@k does not show the calculation how it is actually done in code: ```python def _calculate_average_precision(self, json_responses: t.List[t.Dict]) -> float: score = np.nan json_responses = [ item if isinstance(item, dict) else {} for item in json_responses ] verdict_list = [ int("1" == resp.get("verdict", "").strip()) if resp.get("verdict") else np.nan for resp in json_responses ] denominator = sum(verdict_list) + 1e-10 numerator = sum( [ (sum(verdict_list[: i + 1]) / (i + 1)) * verdict_list[i] for i in range(len(verdict_list)) ] ) score = numerator / denominator if np.isnan(score): logger.warning( "Invalid response format. Expected a list of dictionaries with keys 'verdict'" ) return score ``` There a weighted precision@k is used based on the relevance indicator: ```python (sum(verdict_list[: i + 1]) / (i + 1)) * verdict_list[i] for i in range(len(verdict_list)) ``` Otherwise results greater 1 would be possible e.g. for verdict_list = [1,0]. You can easily verify this with this examples: ```python from ragas.metrics import ContextPrecision from datasets import Dataset from ragas import evaluate import os os.environ["OPENAI_API_KEY"] = "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" os.environ["RAGAS_DO_NOT_TRACK"] = "true" context_precision = ContextPrecision() questions = [ "Where is France and what is it’s capital?", "Where is France and what is it’s capital?", "Where is France and what is it’s capital?", "Where is France and what is it’s capital?", ] ground_truths = [ "France is in Western Europe and its capital is Paris.", "France is in Western Europe and its capital is Paris.", "France is in Western Europe and its capital is Paris.", "France is in Western Europe and its capital is Paris.", ] contexts =[ [ # all bad "The country is also renowned for its wines and sophisticated cuisine. Lascaux’s ancient cave drawings, Lyon’s Roman theater and", "The country is also renowned for its wines and sophisticated cuisine. Lascaux’s ancient cave drawings, Lyon’s Roman theater and", ], [ # wrong order "The country is also renowned for its wines and sophisticated cuisine. Lascaux’s ancient cave drawings, Lyon’s Roman theater and", "France, in Western Europe, encompasses medieval cities, alpine villages and Mediterranean beaches. Paris, its capital, is famed for its fashion houses, classical art museums including the Louvre and monuments like the Eiffel Tower", ], [ # right order "France, in Western Europe, encompasses medieval cities, alpine villages and Mediterranean beaches. Paris, its capital, is famed for its fashion houses, classical art museums including the Louvre and monuments like the Eiffel Tower", "The country is also renowned for its wines and sophisticated cuisine. Lascaux’s ancient cave drawings, Lyon’s Roman theater and", ], [ # all good "France, in Western Europe, encompasses medieval cities, alpine villages and Mediterranean beaches. Paris, its capital, is famed for its fashion houses, classical art museums including the Louvre and monuments like the Eiffel Tower", "France, in Western Europe, encompasses medieval cities, alpine villages and Mediterranean beaches. Paris, its capital, is famed for its fashion houses, classical art museums including the Louvre and monuments like the Eiffel Tower", ] ] data = { "question": questions, "contexts": contexts, "ground_truth": ground_truths } # Convert dict to dataset dataset = Dataset.from_dict(data) dataset result = evaluate( dataset = dataset, metrics=[ context_precision, ], ) print(result.to_pandas()) ``` ``` question ... context_precision 0 Where is France and what is it’s capital? ... 0.0 1 Where is France and what is it’s capital? ... 0.5 2 Where is France and what is it’s capital? ... 1.0 3 Where is France and what is it’s capital? ... 1.0 ``` 3) The `ground_truth` is used as well to calculate the metric.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Precision@1 = 0/1 != 1. It should be 0.
This PR fixes the calculation of Precision@1 and subsequently
\text{Context Precision} = {\text{(0+0.5)} \over \text{2}} = 0.25