Skip to content

~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.

10.4 KB
Raw Blame
"""
Behavioral tests for benchmarks/metrics/calculation module.

Tests pure logic functions: calculate_resource_metrics, calculate_combined_score,
and calculate_metrics with temp file.
"""

import json
import os
import tempfile

import pytest


class TestCalculateResourceMetrics:
    """Tests for calculate_resource_metrics() function."""

    def test_default_config(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_resource_metrics,
        )

        result = calculate_resource_metrics({})
        assert "resource_score" in result
        assert "estimated_complexity" in result

    def test_resource_score_between_0_and_1(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_resource_metrics,
        )

        result = calculate_resource_metrics(
            {"iterations": 5, "questions_per_iteration": 5}
        )
        assert 0.0 <= result["resource_score"] <= 1.0

    def test_higher_iterations_lower_score(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_resource_metrics,
        )

        low = calculate_resource_metrics(
            {"iterations": 1, "questions_per_iteration": 1}
        )
        high = calculate_resource_metrics(
            {"iterations": 10, "questions_per_iteration": 10}
        )
        assert low["resource_score"] > high["resource_score"]

    def test_complexity_increases_with_iterations(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_resource_metrics,
        )

        low = calculate_resource_metrics({"iterations": 1})
        high = calculate_resource_metrics({"iterations": 5})
        assert high["estimated_complexity"] > low["estimated_complexity"]

    def test_max_results_affects_complexity(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_resource_metrics,
        )

        low = calculate_resource_metrics({"max_results": 10})
        high = calculate_resource_metrics({"max_results": 200})
        assert high["estimated_complexity"] > low["estimated_complexity"]

    def test_default_values_used_when_not_specified(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_resource_metrics,
        )

        result = calculate_resource_metrics({})
        # Default: iterations=2, questions=2, max_results=50
        # complexity = 2 * 2 * (50/50) = 4.0
        assert result["estimated_complexity"] == pytest.approx(4.0)


class TestCalculateCombinedScore:
    """Tests for calculate_combined_score() function."""

    def test_default_weights(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_combined_score,
        )

        metrics = {
            "quality": {"quality_score": 1.0},
            "speed": {"speed_score": 1.0},
            "resource": {"resource_score": 1.0},
        }
        score = calculate_combined_score(metrics)
        assert score == pytest.approx(1.0)

    def test_zero_metrics(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_combined_score,
        )

        metrics = {
            "quality": {"quality_score": 0.0},
            "speed": {"speed_score": 0.0},
            "resource": {"resource_score": 0.0},
        }
        score = calculate_combined_score(metrics)
        assert score == 0.0

    def test_quality_weighted_most(self):
        """Quality has default weight 0.6, so it dominates the score."""
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_combined_score,
        )

        quality_only = {
            "quality": {"quality_score": 1.0},
            "speed": {"speed_score": 0.0},
            "resource": {"resource_score": 0.0},
        }
        speed_only = {
            "quality": {"quality_score": 0.0},
            "speed": {"speed_score": 1.0},
            "resource": {"resource_score": 0.0},
        }
        q_score = calculate_combined_score(quality_only)
        s_score = calculate_combined_score(speed_only)
        assert q_score > s_score

    def test_custom_weights(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_combined_score,
        )

        metrics = {
            "quality": {"quality_score": 0.5},
            "speed": {"speed_score": 0.8},
        }
        score = calculate_combined_score(
            metrics, weights={"quality": 0.5, "speed": 0.5}
        )
        # (0.5 * 0.5 + 0.8 * 0.5) / 1.0 = 0.65
        assert score == pytest.approx(0.65)

    def test_zero_total_weight_returns_zero(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_combined_score,
        )

        score = calculate_combined_score({}, weights={})
        assert score == 0.0

    def test_empty_metrics(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_combined_score,
        )

        score = calculate_combined_score({})
        assert score == 0.0

    def test_partial_metrics(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_combined_score,
        )

        metrics = {"quality": {"quality_score": 0.8}}
        score = calculate_combined_score(metrics)
        # Default weights: quality=0.6, speed=0.3, resource=0.1
        # Only quality contributes: 0.8 * 0.6 = 0.48
        assert score == pytest.approx(0.48)


class TestCalculateMetrics:
    """Tests for calculate_metrics() function with temp files."""

    def test_empty_file_returns_error(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_metrics,
        )

        with tempfile.NamedTemporaryFile(
            mode="w", suffix=".jsonl", delete=False
        ) as f:
            f.write("")
            f.flush()
            result = calculate_metrics(f.name)
        os.unlink(f.name)
        assert "error" in result

    def test_single_correct_result(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_metrics,
        )

        with tempfile.NamedTemporaryFile(
            mode="w", suffix=".jsonl", delete=False
        ) as f:
            f.write(
                json.dumps({"is_correct": True, "processing_time": 5.0}) + "\n"
            )
            f.flush()
            result = calculate_metrics(f.name)
        os.unlink(f.name)
        assert result["accuracy"] == 1.0
        assert result["total_examples"] == 1
        assert result["correct"] == 1

    def test_mixed_results_accuracy(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_metrics,
        )

        with tempfile.NamedTemporaryFile(
            mode="w", suffix=".jsonl", delete=False
        ) as f:
            f.write(json.dumps({"is_correct": True}) + "\n")
            f.write(json.dumps({"is_correct": False}) + "\n")
            f.write(json.dumps({"is_correct": True}) + "\n")
            f.flush()
            result = calculate_metrics(f.name)
        os.unlink(f.name)
        assert result["accuracy"] == pytest.approx(2 / 3)

    def test_processing_time_average(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_metrics,
        )

        with tempfile.NamedTemporaryFile(
            mode="w", suffix=".jsonl", delete=False
        ) as f:
            f.write(json.dumps({"processing_time": 10.0}) + "\n")
            f.write(json.dumps({"processing_time": 20.0}) + "\n")
            f.flush()
            result = calculate_metrics(f.name)
        os.unlink(f.name)
        assert result["average_processing_time"] == pytest.approx(15.0)

    def test_error_count(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_metrics,
        )

        with tempfile.NamedTemporaryFile(
            mode="w", suffix=".jsonl", delete=False
        ) as f:
            f.write(json.dumps({"error": "something failed"}) + "\n")
            f.write(json.dumps({"is_correct": True}) + "\n")
            f.flush()
            result = calculate_metrics(f.name)
        os.unlink(f.name)
        assert result["error_count"] == 1
        assert result["error_rate"] == pytest.approx(0.5)

    def test_category_metrics(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_metrics,
        )

        with tempfile.NamedTemporaryFile(
            mode="w", suffix=".jsonl", delete=False
        ) as f:
            f.write(
                json.dumps({"is_correct": True, "category": "science"}) + "\n"
            )
            f.write(
                json.dumps({"is_correct": False, "category": "science"}) + "\n"
            )
            f.write(
                json.dumps({"is_correct": True, "category": "history"}) + "\n"
            )
            f.flush()
            result = calculate_metrics(f.name)
        os.unlink(f.name)
        assert "categories" in result
        assert result["categories"]["science"]["accuracy"] == pytest.approx(0.5)
        assert result["categories"]["history"]["accuracy"] == 1.0

    def test_nonexistent_file_returns_error(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_metrics,
        )

        result = calculate_metrics("/nonexistent/path/file.jsonl")
        assert "error" in result

    def test_confidence_average(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_metrics,
        )

        with tempfile.NamedTemporaryFile(
            mode="w", suffix=".jsonl", delete=False
        ) as f:
            f.write(json.dumps({"confidence": 80}) + "\n")
            f.write(json.dumps({"confidence": 60}) + "\n")
            f.flush()
            result = calculate_metrics(f.name)
        os.unlink(f.name)
        assert result["average_confidence"] == pytest.approx(70.0)

    def test_result_has_timestamp(self):
        from local_deep_research.benchmarks.metrics.calculation import (
            calculate_metrics,
        )

        with tempfile.NamedTemporaryFile(
            mode="w", suffix=".jsonl", delete=False
        ) as f:
            f.write(json.dumps({"is_correct": True}) + "\n")
            f.flush()
            result = calculate_metrics(f.name)
        os.unlink(f.name)
        assert "timestamp" in result