Table of Contents

Class AgentRecommendation<T, TInput, TOutput>

Namespace
AiDotNet.Models
Assembly
AiDotNet.dll

Stores the AI agent's analysis results and recommendations after examining your data and model configuration.

public class AgentRecommendation<T, TInput, TOutput>

Type Parameters

T

The numeric type used for model parameters and calculations (typically float or double).

TInput

The type of input data (e.g., Matrix<double>, Vector<float>).

TOutput

The type of output predictions (e.g., Vector<double>, double[]).

Inheritance
AgentRecommendation<T, TInput, TOutput>
Inherited Members

Remarks

This class encapsulates all the insights, recommendations, and reasoning provided by the AI agent during the model building process. It contains the results of various agent operations such as data analysis, model selection, hyperparameter tuning, and feature analysis. Each property corresponds to a specific agent capability that can be enabled through AgentAssistanceOptions. Properties will only be populated if their corresponding feature was enabled; otherwise, they remain null. This class is returned as part of the AiModelResult and provides valuable insights into why the agent made specific recommendations.

For Beginners: This class holds all the advice and recommendations the AI agent provides about your machine learning model.

Think of it as a detailed report from an AI consultant that contains:

  • Analysis of your data: What patterns, issues, or characteristics the AI found
  • Model recommendations: Which model to use and why
  • Setting suggestions: What hyperparameters to use and why those values make sense
  • Feature insights: Which input variables matter most and how to improve them
  • Reasoning explanations: The AI's thought process behind each recommendation

How it works:

  • Only the sections you enabled in AgentAssistanceOptions will be filled in
  • Each property contains detailed, human-readable explanations
  • You can read these insights to understand the AI's recommendations
  • The reasoning helps you learn about machine learning best practices

For example, if you enabled ModelSelection and HyperparameterTuning, you might get:

  • SuggestedModelType: RandomForest
  • ModelSelectionReasoning: "Random Forest is recommended because your dataset has 10,000 samples with complex non-linear relationships. It will handle the outliers better than linear models and is less likely to overfit than a deep neural network given your data size."
  • SuggestedHyperparameters: {"n_trees": 100, "max_depth": 15, "min_samples_split": 10}
  • TuningReasoning: "These hyperparameters balance model complexity with generalization..."

This information helps you understand not just what to do, but why it's the right approach for your specific problem.

Properties

CompressionReasoning

Gets or sets the AI agent's detailed explanation for the compression recommendation.

public string? CompressionReasoning { get; set; }

Property Value

string

A detailed text explanation of the compression selection rationale, or null if not available.

Remarks

This property explains why the agent recommended a specific compression technique, including analysis of the model structure and expected compression-accuracy tradeoffs.

For Beginners: This explains why a particular compression technique was chosen.

For example: "Deep Compression (pruning + quantization + Huffman) is recommended because:

  1. Your model has many fully-connected layers with high redundancy
  2. 92% of weights have magnitude less than 0.1 (good pruning candidates)
  3. Expected compression: 35-50x with less than 1% accuracy loss
  4. Target platform (mobile) requires small model size

Alternative considered: Weight Clustering alone would give 8x compression, but Deep Compression's multi-stage approach will achieve better results for your deployment constraints."

DataAnalysis

Gets or sets the AI agent's analysis of your training data, including patterns, issues, and characteristics.

public string? DataAnalysis { get; set; }

Property Value

string

A detailed text description of data analysis insights, or null if data analysis was not enabled.

Remarks

This property contains the agent's comprehensive analysis of your training data when EnableDataAnalysis is turned on. The analysis includes observations about data distributions, outliers, missing values, feature correlations, class imbalances, and potential data quality issues. The agent provides this information in natural language with specific examples and statistics from your data. This property is null if data analysis was not enabled in AgentAssistanceOptions.

For Beginners: This contains what the AI discovered when examining your data.

What you'll find here:

  • Unusual patterns or anomalies in your data
  • Statistics about your features and target variable
  • Warnings about potential problems (missing data, outliers, etc.)
  • Observations about feature relationships and correlations
  • Suggestions for data preprocessing or cleaning

For example, you might see: "Your dataset contains 5,000 samples with 12 features. The 'price' feature shows significant right skew with some extreme outliers above $2M. Features 'sqft' and 'bedrooms' are highly correlated (r=0.85). About 8% of 'year_built' values are missing. Consider log-transforming 'price' to handle the skew, and either remove or impute the missing values before training."

This helps you understand your data better and catch issues before they affect your model's performance.

ExpectedCompressionMetrics

Gets or sets the expected compression metrics after applying the recommended technique.

public string? ExpectedCompressionMetrics { get; set; }

Property Value

string

Predicted compression metrics, or null if not available.

Remarks

This property contains the agent's prediction of what compression metrics you can expect if you apply the recommended compression technique. These are estimates based on model analysis.

For Beginners: This shows what to expect from compression before you apply it.

Typical information includes:

  • Expected compression ratio (e.g., "20-30x smaller")
  • Expected accuracy loss (e.g., "less than 1%")
  • Expected inference speedup (e.g., "2-3x faster")

These are estimates - actual results depend on your specific model and data.

FeatureRecommendations

Gets or sets the AI agent's recommendations about feature selection, importance, and engineering.

public string? FeatureRecommendations { get; set; }

Property Value

string

A detailed text description of feature analysis and recommendations, or null if feature analysis was not enabled.

Remarks

This property contains the agent's analysis of your input features when EnableFeatureAnalysis is turned on. It includes rankings of feature importance, identification of redundant or unhelpful features, suggestions for new features to create through feature engineering, recommendations for feature transformations, and insights about feature interactions. The agent explains which features are most predictive, which could be removed without hurting performance, and how you might improve your feature set. This property is null if feature analysis was not enabled in AgentAssistanceOptions.

For Beginners: This tells you which input variables (features) matter most and how to improve them.

What you'll learn:

  • Which features are most important for predictions
  • Which features aren't helping and could be removed
  • Ideas for new features to create (feature engineering)
  • Transformations that might improve feature usefulness
  • How features interact with each other

For example, when predicting house prices: "Feature Analysis:

Most Important Features:

  1. square_footage (importance: 0.45) - Strongest predictor of price
  2. location_score (importance: 0.28) - Second most influential
  3. year_built (importance: 0.12) - Moderate impact

Low-Impact Features:

  • exterior_color (importance: 0.01) - Consider removing
  • HOA_name (importance: 0.00) - Not predictive, remove

Feature Engineering Suggestions:

  • Create 'price_per_sqft' = price / square_footage (useful for normalization)
  • Create 'age' = current_year - year_built (more intuitive than year_built)
  • Consider binning 'year_built' into decades for non-linear patterns

The strong correlation between square_footage and bedrooms (r=0.85) suggests multicollinearity. Consider using just one or creating a composite feature."

This helps you focus on important features and improve your model's performance through better feature engineering.

ModelSelectionReasoning

Gets or sets the AI agent's detailed explanation for why it recommended the suggested model type.

public string? ModelSelectionReasoning { get; set; }

Property Value

string

A detailed text explanation of the model selection rationale, or null if model selection was not enabled.

Remarks

This property contains the agent's natural language explanation of why it recommended the SuggestedModelType. The reasoning describes how the model's characteristics align with your data properties, explains what makes this model a good fit, discusses potential alternatives that were considered, and may warn about limitations or considerations to keep in mind. This educational explanation helps you understand the decision-making process rather than just receiving a recommendation blindly. This property is null if model selection was not enabled.

For Beginners: This explains why the AI chose the recommended model - the "why" behind the "what."

The reasoning typically covers:

  • Why this model suits your specific data characteristics
  • What advantages this model has for your problem
  • What alternatives were considered and why they weren't chosen
  • Any limitations or caveats to be aware of
  • How this model compares to others for your use case

For example: "Random Forest is recommended for your dataset because:

  1. Your 10,000 samples provide enough data for an ensemble method
  2. The non-linear relationships in your features need a flexible model
  3. Random Forest handles the outliers in 'price' better than linear models
  4. It's less prone to overfitting than a single decision tree
  5. It provides feature importance scores which will be valuable for your analysis

While Neural Networks could also work, Random Forest will likely train faster and be easier to interpret for your business stakeholders."

This helps you learn about model selection while getting personalized advice for your situation.

ReasoningTrace

Gets or sets the complete reasoning trace showing the AI agent's thought process across all operations.

public string? ReasoningTrace { get; set; }

Property Value

string

A comprehensive text record of the agent's reasoning steps and decisions, or null if no agent operations were performed.

Remarks

This property contains a chronological trace of all reasoning steps performed by the AI agent during the model building process. It combines insights from all enabled agent operations into a single comprehensive narrative, showing how the agent analyzed your data, considered different options, made decisions, and arrived at its recommendations. This is particularly useful for understanding the complete picture of the agent's analysis and for debugging or learning purposes. This property is null if no agent assistance features were enabled.

For Beginners: This is the complete record of everything the AI thought about and decided during the model building process.

It's like reading the AI's notebook where it:

  • Recorded observations about your data
  • Explained its decision-making process
  • Showed how different pieces of analysis connect
  • Documented why it made each recommendation

This is the most detailed view of the agent's work. While the other properties give you specific pieces (data analysis, model choice, hyperparameters, features), this property shows the complete story of how all those pieces fit together.

Use this when you want to:

  • Understand the full picture of the AI's analysis
  • Learn about the machine learning process
  • Debug unexpected recommendations
  • See how the AI connects different insights

For example, it might show: "Step 1: Analyzed data distribution and found right-skewed target variable... Step 2: Based on skewness and dataset size, considered RandomForest and GradientBoosting... Step 3: Selected RandomForest because it's more robust to outliers... Step 4: Tuned hyperparameters considering the dataset size of 5,000 samples... Step 5: Identified square_footage as the most important feature due to high correlation with target..."

This comprehensive view helps you learn how an expert approaches machine learning problems.

SuggestedCompressionParameters

Gets or sets the recommended hyperparameters for the compression technique.

public Dictionary<string, object>? SuggestedCompressionParameters { get; set; }

Property Value

Dictionary<string, object>

A dictionary mapping parameter names to recommended values, or null if not available.

Remarks

This property contains the specific hyperparameter values recommended for the suggested compression technique. These are tuned based on the model characteristics and constraints.

For Beginners: These are the settings to use with the recommended compression technique.

For example, for Deep Compression:

  • "pruningSparsity": 0.9 (remove 90% of smallest weights)
  • "numClusters": 32 (use 5-bit quantization)
  • "huffmanPrecision": 4 (4 decimal places for encoding)

These values were chosen based on your model's characteristics to balance compression and accuracy.

SuggestedCompressionType

Gets or sets the AI agent's recommended compression technique for model deployment.

public CompressionType? SuggestedCompressionType { get; set; }

Property Value

CompressionType?

The suggested compression type, or null if compression analysis was not enabled.

Remarks

This property contains the specific compression technique that the AI agent recommends based on the model architecture, deployment constraints, and accuracy requirements. The recommendation considers factors like model size, layer types, and target platform.

For Beginners: This tells you which compression method to use to make your model smaller.

Common recommendations include:

  • SparsePruning: Remove small weights (good for large, dense models)
  • WeightClustering: Group similar weights (good for moderate compression)
  • DeepCompression: Combine multiple techniques (best for maximum compression)

The agent chooses based on:

  • Your model's architecture (CNNs vs fully-connected)
  • Your accuracy tolerance (how much accuracy loss is acceptable)
  • Your deployment target (mobile, cloud, edge devices)

SuggestedHyperparameters

Gets or sets the AI agent's recommended hyperparameter values for the selected model.

public Dictionary<string, object>? SuggestedHyperparameters { get; set; }

Property Value

Dictionary<string, object>

A dictionary mapping hyperparameter names to their recommended values, or null if hyperparameter tuning was not enabled.

Remarks

This property contains a dictionary of hyperparameter names and their recommended values for your selected model. The keys are hyperparameter names (like "learning_rate", "max_depth", "n_estimators") and the values are the suggested settings. These recommendations are tailored to your specific data characteristics and model type. The agent considers factors like dataset size, feature dimensionality, and problem complexity when suggesting these values. This property is null if hyperparameter tuning was not enabled in AgentAssistanceOptions. See TuningReasoning for detailed explanations of why these values were chosen.

For Beginners: This contains the recommended settings for your model - like tuning knobs on a machine.

Hyperparameters control how your model learns:

  • How fast it learns (learning_rate)
  • How complex it can get (max_depth, n_layers)
  • How many parts it has (n_estimators, n_neurons)
  • How it handles data (batch_size, regularization)

For example, if using Random Forest, you might get: { "n_estimators": 100, // Use 100 trees "max_depth": 15, // Each tree can be 15 levels deep "min_samples_split": 10, // Need at least 10 samples to split "max_features": "sqrt" // Consider sqrt(n) features per split }

These aren't random guesses - the AI chose these values specifically for your data size and characteristics. Check TuningReasoning to understand why each value was selected.

SuggestedModelType

Gets or sets the AI agent's recommended model type for your problem.

public ModelType? SuggestedModelType { get; set; }

Property Value

ModelType?

A ModelType enum value representing the recommended model, or null if model selection was not enabled.

Remarks

This property contains the specific model type that the AI agent recommends based on your data characteristics, problem complexity, and dataset size. The recommendation is made by analyzing factors such as linearity, feature dimensionality, sample size, presence of outliers, and computational constraints. This property is null if model selection was not enabled in AgentAssistanceOptions. See the ModelSelectionReasoning property for the explanation of why this model was chosen.

For Beginners: This tells you which machine learning model the AI thinks you should use.

The AI chooses based on:

  • Your data size (small, medium, or large dataset)
  • Complexity of patterns in your data
  • Type of relationships (linear, non-linear, etc.)
  • Presence of outliers or unusual data points

For example, the AI might recommend:

  • SimpleRegression for small datasets with clear linear relationships
  • RandomForest for medium-sized datasets with complex patterns
  • NeuralNetworkRegression for large datasets with very complex relationships

Always read the ModelSelectionReasoning property to understand why this model was chosen for your specific situation.

TuningReasoning

Gets or sets the AI agent's detailed explanation for the recommended hyperparameter values.

public string? TuningReasoning { get; set; }

Property Value

string

A detailed text explanation of the hyperparameter tuning rationale, or null if hyperparameter tuning was not enabled.

Remarks

This property contains the agent's natural language explanation of why it chose the specific hyperparameter values in SuggestedHyperparameters. For each major hyperparameter, the reasoning explains what the parameter controls, how it affects model behavior, why the suggested value is appropriate for your data, and what trade-offs were considered. This educational explanation helps you understand hyperparameter tuning principles rather than just applying recommended values blindly. This property is null if hyperparameter tuning was not enabled.

For Beginners: This explains why the AI chose these specific hyperparameter values.

For each important setting, you'll learn:

  • What the hyperparameter does and why it matters
  • Why this specific value is good for your data
  • What happens if you increase or decrease the value
  • How it balances different trade-offs (speed vs accuracy, simplicity vs power)

For example: "Hyperparameter recommendations for your Random Forest:

n_estimators=100: Using 100 trees provides a good balance. More trees improve accuracy but with diminishing returns beyond 100 for your dataset size. Fewer trees might underfit your complex patterns.

max_depth=15: Limiting tree depth prevents overfitting. Your dataset's complexity needs depth to capture patterns, but unlimited depth would memorize noise in your 5,000 training samples.

min_samples_split=10: Requiring 10 samples before splitting helps avoid creating overly specific rules based on just a few data points, which improves generalization to new data.

These values are tuned for your specific dataset size and complexity. If you have more data later, you could increase max_depth to capture more detailed patterns."

This helps you understand hyperparameter tuning and make informed adjustments if needed.

Edit this page