Statistical Test Results
| Test |
#{format_test_name(result.test)} |
| Test Statistic |
#{format_number(result.statistic)} |
| P-value |
#{format_p_value(result.p_value)}
|
#{format_effect_size_html(result.effect_size)}
#{format_confidence_interval_html(result.confidence_interval)}
Interpretation
#{result.interpretation || "No interpretation available"}
#{format_metadata_html(result.metadata)}
"""
|> String.trim()
end
@doc """
Export experiment results to comprehensive report.
"""
def experiment_to_markdown(experiment_result) do
case experiment_result.experiment_type do
:ab_test -> ab_test_report(experiment_result)
:ablation -> ablation_report(experiment_result)
:hyperparameter_sweep -> sweep_report(experiment_result)
end
end
defp ab_test_report(exp) do
"""
# A/B Test Report: #{exp.name}
## Summary
#{if exp.significant?, do: "✓", else: "✗"} **Result**: #{if exp.significant?, do: "Significant difference detected", else: "No significant difference"}
## Statistics
- **P-value**: #{format_p_value(exp.p_value)}
- **Test Used**: #{format_test_name(exp.test_used)}
- **Effect Size**: #{exp.effect_size.interpretation} (Cohen's d = #{format_number(exp.effect_size.cohens_d)})
- **Statistical Power**: #{Float.round(exp.power * 100, 1)}%
## Sample Sizes
- Control: #{exp.sample_sizes.control}
- Treatment: #{exp.sample_sizes.treatment}
## Performance Metrics
| Group | Mean | Difference |
|-------|------|------------|
| Control | #{format_number(exp.means.control)} | - |
| Treatment | #{format_number(exp.means.treatment)} | #{format_diff(exp.means.difference)} |
## Confidence Interval (95%)
#{format_ci_range(exp.confidence_interval)}
## Interpretation
#{exp.interpretation}
## Recommendation
#{exp.recommendation}
"""
|> String.trim()
end
defp ablation_report(exp) do
"""
# Ablation Study: #{exp.component_name}
## Summary
#{if exp.significant_impact?, do: "✓ Significant impact", else: "✗ No significant impact"}
## Statistics
- **P-value**: #{format_p_value(exp.p_value)}
- **Effect Size**: #{exp.effect_size.interpretation} (Cohen's d = #{format_number(exp.effect_size.cohens_d)})
## Performance Impact
- **Absolute Drop**: #{format_number(exp.performance_drop.absolute)}
- **Percent Drop**: #{format_number(exp.performance_drop.percent)}%
## Performance Comparison
| Configuration | Mean Performance |
|--------------|------------------|
| Baseline (with component) | #{format_number(exp.means.baseline)} |
| Without component | #{format_number(exp.means.without_component)} |
## Interpretation
#{exp.interpretation}
"""
|> String.trim()
end
defp sweep_report(exp) do
"""
# Hyperparameter Sweep Results
## Summary
**Configurations Tested**: #{exp.configurations_tested}
**Best Configuration**: #{exp.best_configuration.name} (mean = #{format_number(exp.best_configuration.mean)})
## Omnibus Test
- **Significant Differences**: #{if exp.omnibus_test.significant?, do: "Yes", else: "No"}
- **P-value**: #{format_p_value(exp.omnibus_test.p_value)}
- **Test Used**: #{format_test_name(exp.omnibus_test.test_used)}
- **Effect Size**: #{exp.omnibus_test.effect_size.interpretation}
## Configuration Means
| Configuration | Mean Performance |
|--------------|------------------|
#{format_config_means(exp.configuration_means)}
## Pairwise Comparisons
#{format_pairwise_comparisons(exp.pairwise_comparisons)}
## Interpretation
#{exp.interpretation}
"""
|> String.trim()
end
# Formatting helpers
defp format_test_name(:welch_t_test), do: "Welch's t-test"
defp format_test_name(:student_t_test), do: "Student's t-test"
defp format_test_name(:paired_t_test), do: "Paired t-test"
defp format_test_name(:anova), do: "One-way ANOVA"
defp format_test_name(:mann_whitney), do: "Mann-Whitney U test"
defp format_test_name(:wilcoxon_signed_rank), do: "Wilcoxon signed-rank test"
defp format_test_name(:kruskal_wallis), do: "Kruskal-Wallis test"
defp format_test_name(other), do: to_string(other)
defp format_number(nil), do: "N/A"
defp format_number(n) when is_float(n), do: Float.round(n, 4) |> to_string()
defp format_number(n), do: to_string(n)
defp format_p_value(p) when p < 0.001, do: "< 0.001"
defp format_p_value(p), do: format_number(p)
defp format_p_value_latex(p) when p < 0.001, do: "$< 0.001$"
defp format_p_value_latex(p), do: "$#{format_number(p)}$"
defp format_diff(d) when d > 0, do: "+#{format_number(d)}"
defp format_diff(d), do: format_number(d)
defp format_ci_range({lower, upper}) do
"[#{format_number(lower)}, #{format_number(upper)}]"
end
defp format_ci_range(nil), do: "N/A"
defp format_effect_size_md(nil), do: ""
defp format_effect_size_md(effect) when is_map(effect) do
"**Effect Size**: #{format_effect_size_value(effect)}\n"
end
defp format_effect_size_value(%{cohens_d: d}), do: "Cohen's d = #{format_number(d)}"
defp format_effect_size_value(%{eta_squared: eta}), do: "η² = #{format_number(eta)}"
defp format_effect_size_value(%{rank_biserial: r}), do: "Rank-biserial r = #{format_number(r)}"
defp format_effect_size_value(_), do: "N/A"
defp format_confidence_interval_md(nil), do: ""
defp format_confidence_interval_md(ci) do
"**95% Confidence Interval**: #{format_ci_range(ci)}\n"
end
defp format_effect_size_latex(nil), do: ""
defp format_effect_size_latex(effect) when is_map(effect) do
"\\textbf{Effect Size} & #{format_effect_size_value(effect)} \\\\\n"
end
defp format_confidence_interval_latex(nil), do: ""
defp format_confidence_interval_latex(ci) do
"\\textbf{95\\% CI} & #{format_ci_range(ci)} \\\\\n"
end
defp format_effect_size_html(nil), do: ""
defp format_effect_size_html(effect) when is_map(effect) do
"""