💡 Model Recommendations

Based on the analysis of all tested configurations, here are the recommended models for different scenarios:

{% for rec in recommendations %}

{% if rec.metrics.accuracy %}

Accuracy: {{ "%.4f" | format(rec.metrics.accuracy) }}

{% endif %} {% if rec.metrics.f1_score %}

F1: {{ "%.4f" | format(rec.metrics.f1_score) }}

{% endif %} {% if rec.metrics.training_time %}

Time: {{ "%.2f" | format(rec.metrics.training_time) }}s

{% endif %} {% if rec.metrics.complexity %}

Complexity: {{ "%.2f" | format(rec.metrics.complexity) }}

{% endif %} {% if rec.metrics.efficiency %}

Efficiency: {{ "%.3f" | format(rec.metrics.efficiency) }}

{% endif %} {% if rec.metrics.improvement %}

Improvement: +{{ "%.4f" | format(rec.metrics.improvement) }}

{% endif %} {% if rec.metrics.production_score %}

Prod Score: {{ "%.3f" | format(rec.metrics.production_score) }}

{% endif %}

{% endfor %}

📋 Decision Matrix

Choose the right model based on your priorities:

Priority	Recommended Model	Key Advantage	Trade-off
Maximum Accuracy	{{ best_model.model_type }} (T={{ best_model.temperature }}, α={{ best_model.alpha }})	Highest test accuracy	May have higher complexity
Fast Inference	Look for low complexity models	Quick predictions	Potentially lower accuracy
Resource Constrained	Models with lowest complexity score	Minimal memory/compute needs	Reduced performance
Production Ready	Balanced score models	Good accuracy with reasonable resources	Not the absolute best in any metric
Experimentation	Fast training models	Quick iteration cycles	May not be optimal for deployment

Validate the recommended model on a holdout dataset to confirm performance.
Fine-tune hyperparameters around the optimal ranges identified (Temperature: {{ "%.2f-%.2f" | format( report_data.hyperparameter_analysis.optimal_ranges.temperature.min, report_data.hyperparameter_analysis.optimal_ranges.temperature.max ) }}, Alpha: {{ "%.2f-%.2f" | format( report_data.hyperparameter_analysis.optimal_ranges.alpha.min, report_data.hyperparameter_analysis.optimal_ranges.alpha.max ) }}).
Consider ensemble methods combining multiple distilled models if accuracy is critical.
Evaluate on production data to ensure the model generalizes well.
Monitor performance after deployment and retrain if necessary.