|
| 1 | +# Represents the evaluation results for a single search query |
| 2 | +struct QueryResult |
| 3 | + query::String |
| 4 | + precision::Float64 |
| 5 | + recall::Float64 |
| 6 | + f1::Float64 |
| 7 | + expected::Vector{String} |
| 8 | + actual::Vector{String} |
| 9 | + # Raw integer values used in calculations |
| 10 | + relevant_count::Int # Number of relevant documents found |
| 11 | + total_retrieved::Int # Total number of documents retrieved |
| 12 | + total_relevant::Int # Total number of relevant documents |
| 13 | +end |
| 14 | + |
| 15 | +# Aggregates evaluation results across multiple search queries |
| 16 | +struct EvaluationResults |
| 17 | + individual_results::Vector{QueryResult} |
| 18 | + average_precision::Float64 |
| 19 | + average_recall::Float64 |
| 20 | + average_f1_score::Float64 |
| 21 | + # Raw integer values for overall evaluation |
| 22 | + total_relevant_found::Int # Total number of relevant documents found across all queries |
| 23 | + total_documents_retrieved::Int # Total number of documents retrieved across all queries |
| 24 | + total_relevant_documents::Int # Total number of relevant documents across all queries |
| 25 | +end |
| 26 | + |
| 27 | +# Calculates precision for search results against expected documents |
| 28 | +# Precision = (relevant documents found) / (total documents retrieved) |
| 29 | +# Returns precision score, count of relevant documents found, and total documents retrieved |
| 30 | +function calculate_precision(results, expected_docs) |
| 31 | + if isempty(results) |
| 32 | + return 0.0, 0, 0 |
| 33 | + end |
| 34 | + |
| 35 | + relevant_count = length(intersect(results, expected_docs)) |
| 36 | + total_retrieved = length(results) |
| 37 | + |
| 38 | + return relevant_count / total_retrieved, relevant_count, total_retrieved |
| 39 | +end |
| 40 | + |
| 41 | +# Calculates recall for search results against expected documents |
| 42 | +# Recall = (relevant documents found) / (total relevant documents) |
| 43 | +# Measures completeness of the search results - how many of the relevant documents were found |
| 44 | +# Returns recall score, count of relevant documents found, and total relevant documents |
| 45 | +function calculate_recall(results, expected_docs) |
| 46 | + if isempty(expected_docs) |
| 47 | + return 1.0, 0, 0 |
| 48 | + end |
| 49 | + |
| 50 | + found_count = length(intersect(results, expected_docs)) |
| 51 | + total_relevant = length(expected_docs) |
| 52 | + |
| 53 | + return found_count / total_relevant, found_count, total_relevant |
| 54 | +end |
| 55 | + |
| 56 | +# Calculates F1 score from precision and recall values |
| 57 | +# F1 = 2 * (precision * recall) / (precision + recall) |
| 58 | +# Combines precision and recall into a single score, giving equal weight to both metrics |
| 59 | +# Returns 0.0 if both precision and recall are 0 |
| 60 | +function calculate_f1(precision, recall) |
| 61 | + if precision + recall == 0 |
| 62 | + return 0.0 |
| 63 | + end |
| 64 | + |
| 65 | + return 2 * (precision * recall) / (precision + recall) |
| 66 | +end |
| 67 | + |
| 68 | +# Evaluates a single search query using the provided search function |
| 69 | +# Returns a QueryResult containing precision, recall, and F1 metrics |
| 70 | +function evaluate_query(search_function, query::TestQuery) |
| 71 | + results = search_function(query.query) |
| 72 | + |
| 73 | + precision, relevant_count, total_retrieved = calculate_precision(results, query.expected_docs) |
| 74 | + recall, found_count, total_relevant = calculate_recall(results, query.expected_docs) |
| 75 | + f1 = calculate_f1(precision, recall) |
| 76 | + |
| 77 | + return QueryResult( |
| 78 | + query.query, |
| 79 | + precision, |
| 80 | + recall, |
| 81 | + f1, |
| 82 | + query.expected_docs, |
| 83 | + results, |
| 84 | + relevant_count, |
| 85 | + total_retrieved, |
| 86 | + total_relevant |
| 87 | + ) |
| 88 | +end |
| 89 | + |
| 90 | +# Evaluates multiple search queries and aggregates the results |
| 91 | +# Returns an EvaluationResults containing average metrics across all queries |
| 92 | +function evaluate_all(search_function, queries) |
| 93 | + results = [evaluate_query(search_function, q) for q in queries] |
| 94 | + |
| 95 | + avg_precision = mean([r.precision for r in results]) |
| 96 | + avg_recall = mean([r.recall for r in results]) |
| 97 | + avg_f1 = mean([r.f1 for r in results]) |
| 98 | + |
| 99 | + # Calculate total raw values across all queries |
| 100 | + total_relevant_found = sum(r.relevant_count for r in results) |
| 101 | + total_documents_retrieved = sum(r.total_retrieved for r in results) |
| 102 | + total_relevant_documents = sum(r.total_relevant for r in results) |
| 103 | + |
| 104 | + return EvaluationResults( |
| 105 | + results, |
| 106 | + avg_precision, |
| 107 | + avg_recall, |
| 108 | + avg_f1, |
| 109 | + total_relevant_found, |
| 110 | + total_documents_retrieved, |
| 111 | + total_relevant_documents |
| 112 | + ) |
| 113 | +end |
0 commit comments