Table of Contents

Class QuantileRegressionForestsOptions

Namespace
AiDotNet.Models.Options
Assembly
AiDotNet.dll

Configuration options for Quantile Regression Forests, an extension of Random Forests that enables prediction of conditional quantiles rather than just the conditional mean.

public class QuantileRegressionForestsOptions : DecisionTreeOptions
Inheritance
QuantileRegressionForestsOptions
Inherited Members

Remarks

Quantile Regression Forests extend the Random Forest algorithm to provide full conditional distributions instead of just point estimates. While standard Random Forests estimate the conditional mean E(Y|X), Quantile Regression Forests can estimate any conditional quantile Q(a|X) for a ? (0,1), including medians and prediction intervals. This is achieved by keeping track of all target values in the leaf nodes of each tree, rather than just their averages. The algorithm provides a non-parametric way to estimate conditional distributions, making it particularly valuable for problems where uncertainty quantification is important or where the conditional distribution is non-Gaussian, skewed, or heteroscedastic (having non-constant variance). Quantile Regression Forests inherit many advantages of Random Forests, including handling of non-linear relationships, robustness to outliers, and minimal parameter tuning requirements.

For Beginners: Quantile Regression Forests help predict not just a single value, but a range of possible values with their probabilities.

Think about weather forecasting:

  • A regular forecast might say "tomorrow's temperature will be 75°F"
  • But Quantile Regression Forests could tell you:
    • "There's a 10% chance it will be below 70°F"
    • "There's a 50% chance it will be below 75°F" (the median)
    • "There's a 90% chance it will be below 80°F"

What this algorithm does:

  • It builds many decision trees, just like a regular Random Forest
  • But instead of averaging their predictions to get a single answer
  • It keeps track of all possible outcomes and their distributions
  • This lets you understand the uncertainty in your predictions

This is especially useful when:

  • You need to know the range of possible outcomes, not just the average
  • Your data has varying levels of uncertainty in different regions
  • The distribution of possible outcomes is not symmetric
  • Risk assessment is as important as the prediction itself

For example, in financial forecasting, knowing there's a 5% chance of losing $10,000 is very different information than just knowing the average expected return.

This class lets you configure how the forest of trees is built and processed.

Properties

MaxDegreeOfParallelism

Gets or sets the maximum degree of parallelism for tree building.

public int MaxDegreeOfParallelism { get; set; }

Property Value

int

The maximum degree of parallelism, defaulting to the number of processor cores.

Remarks

This parameter controls how many trees can be built in parallel during the training process. Since each tree in the forest can be built independently, Quantile Regression Forests are naturally parallelizable. By default, this value is set to the number of logical processors available on the machine, which often provides a good balance between computation speed and resource utilization. Higher values can improve training speed on machines with many cores, while lower values reduce resource consumption. Setting this to 1 disables parallelism, which may be necessary in memory-constrained environments or when debugging. This parameter affects only training speed, not the model's predictive capabilities.

For Beginners: This setting controls how many trees can be built simultaneously using multiple processor cores.

The default value (Environment.ProcessorCount) means:

  • The algorithm will use all available CPU cores on your machine
  • If you have a 4-core processor, it can build 4 trees at the same time
  • This makes training much faster than building trees one at a time

Think of it like a construction project:

  • Each tree is like a house that needs to be built
  • Each processor core is like a construction team
  • With multiple teams working in parallel, you can build houses much faster
  • But using all your teams means you can't use them for other tasks

You might want a higher value:

  • This setting automatically uses all available cores, so increasing it beyond your core count won't help

You might want a lower value (like 2 or 1):

  • When you want to leave CPU resources for other applications
  • When you're running into memory limitations (fewer parallel trees = less memory usage)
  • When you notice your system becoming unresponsive during training

This setting only affects training speed and resource usage - it doesn't change how the model performs once trained. It's a practical consideration rather than a modeling decision.

NumberOfTrees

Gets or sets the number of trees to grow in the forest.

public int NumberOfTrees { get; set; }

Property Value

int

The number of trees, defaulting to 100.

Remarks

This parameter determines how many individual decision trees will be constructed in the Quantile Regression Forest ensemble. Each tree is built using a bootstrap sample of the training data and a random subset of features at each split. A larger number of trees generally improves prediction quality and provides more stable quantile estimates, at the cost of increased computation time and memory usage. The improvement in performance typically diminishes with increasing numbers of trees, with diminishing returns beyond a certain point. For quantile estimation, having more trees can be particularly important to obtain smooth and stable estimates of the conditional distribution.

For Beginners: This setting controls how many different decision trees the algorithm builds.

The default value of 100 means:

  • The algorithm will create 100 different trees
  • Each tree is slightly different because it uses randomly selected data and features
  • The final prediction combines information from all 100 trees

Think of it like getting opinions from multiple experts:

  • Each tree is like a different expert with a slightly different perspective
  • More experts (trees) generally give you more reliable opinions
  • But there's a point where adding more experts doesn't help much more

You might want more trees (like 500 or 1000):

  • When you need more precise quantile estimates
  • When you have a complex problem with many variables
  • When you have enough computational resources available
  • When the stability of quantile estimates is particularly important

You might want fewer trees (like 50 or 20):

  • When you need faster training and prediction times
  • When you have limited computational resources
  • When you're doing initial exploration and don't need optimal performance
  • When your dataset is relatively simple

Adding more trees almost never hurts performance (just computation time), so this is one of the easier parameters to tune: start with 100 and increase if needed.

Edit this page