BayesianDeepLearning

Implements Bayesian Deep Learning techniques for spatial uncertainty quantification in sampling optimization. Provides Monte Carlo dropout, variational inference, and comprehensive uncertainty estimation capabilities.

Details

The BayesianDeepLearning class integrates with the torch framework to provide:

• Monte Carlo dropout for epistemic uncertainty estimation

• Variational inference for parameter uncertainty quantification

• Epistemic, aleatoric, and total uncertainty calculation

• Spatial-aware neural network architectures

Public fields

model

Trained torch neural network model

config

Configuration parameters for BDL

torch_available

Boolean indicating torch availability

uncertainty_cache

Cached uncertainty calculations

Active bindings

model

Trained torch neural network model

config

Configuration parameters for BDL

torch_available

Boolean indicating torch availability

uncertainty_cache

Cached uncertainty calculations

Methods

Public methods

• BayesianDeepLearning$new()

• BayesianDeepLearning$fit_model()

• BayesianDeepLearning$predict_with_uncertainty()

• BayesianDeepLearning$mc_dropout_predict()

• BayesianDeepLearning$variational_inference()

• BayesianDeepLearning$epistemic_uncertainty()

• BayesianDeepLearning$aleatoric_uncertainty()

• BayesianDeepLearning$total_uncertainty()

• BayesianDeepLearning$clone()

---

Method new()

Usage

BayesianDeepLearning$new(config = list())

Arguments

config

Optional configuration list for BDL parameters

Examples

\dontrun{
bdl <- BayesianDeepLearning$new()
bdl_custom <- BayesianDeepLearning$new(config = list(dropout_rate = 0.2))
}

---

Method fit_model()

Usage

BayesianDeepLearning$fit_model(
  field_data,
  existing_samples,
  target_variable = NULL,
  validation_split = 0.2,
  epochs = 100,
  batch_size = 32
)

Arguments

field_data

List containing boundary, covariates, and metadata

existing_samples

Data frame with existing sample locations and values

target_variable

Name of target variable to predict

validation_split

Proportion of data for validation

epochs

Number of training epochs

batch_size

Training batch size

Returns

Fitted BDL model with training metrics

Examples

\dontrun{
model <- bdl$fit_model(
  field_data = field_data,
  existing_samples = samples,
  target_variable = "soil_property",
  epochs = 100
)
}

---

Method predict_with_uncertainty()

Usage

BayesianDeepLearning$predict_with_uncertainty(
  locations,
  n_samples = 100,
  return_samples = FALSE,
  progress_manager = NULL,
  resource_manager = NULL
)

Arguments

locations

Spatial locations for prediction (sf object or data.frame with x,y)

n_samples

Number of Monte Carlo samples for uncertainty estimation

return_samples

Whether to return individual MC samples

progress_manager

Optional ProgressManager instance

resource_manager

Optional ResourceManager instance

Returns

List with predictions and uncertainty estimates

Examples

\dontrun{
predictions <- bdl$predict_with_uncertainty(
  locations = new_locations,
  n_samples = 100
)
}

---

Method mc_dropout_predict()

Usage

BayesianDeepLearning$mc_dropout_predict(
  model,
  data,
  n_iterations = 100,
  dropout_rate = 0.1
)

Arguments

model

Trained neural network model

data

Input data for prediction

n_iterations

Number of MC dropout iterations

dropout_rate

Dropout rate for MC sampling

Returns

Matrix of MC dropout predictions

Examples

\dontrun{
mc_predictions <- bdl$mc_dropout_predict(
  model = fitted_model,
  data = prediction_data,
  n_iterations = 100,
  dropout_rate = 0.1
)
}

---

Method variational_inference()

Usage

BayesianDeepLearning$variational_inference(
  data,
  prior_params = list(),
  n_samples = 1000,
  learning_rate = 0.01
)

Arguments

data

Training data for variational inference

prior_params

Prior parameter specifications

n_samples

Number of variational samples

learning_rate

Learning rate for variational optimization

Returns

Variational inference results with parameter distributions

Examples

\dontrun{
vi_result <- bdl$variational_inference(
  data = training_data,
  prior_params = list(mean = 0, std = 1),
  n_samples = 1000
)
}

---

Method epistemic_uncertainty()

Usage

BayesianDeepLearning$epistemic_uncertainty(predictions)

Arguments

predictions

Prediction results from BDL model

Returns

Vector of epistemic uncertainty estimates

Examples

\dontrun{
epistemic <- bdl$epistemic_uncertainty(predictions)
}

---

Method aleatoric_uncertainty()

Usage

BayesianDeepLearning$aleatoric_uncertainty(predictions)

Arguments

predictions

Prediction results from BDL model

Returns

Vector of aleatoric uncertainty estimates

Examples

\dontrun{
aleatoric <- bdl$aleatoric_uncertainty(predictions)
}

---

Method total_uncertainty()

Usage

BayesianDeepLearning$total_uncertainty(predictions)

Arguments

predictions

Prediction results from BDL model

Returns

Vector of total uncertainty estimates

Examples

\dontrun{
total <- bdl$total_uncertainty(predictions)
}

---

Method clone()

The objects of this class are cloneable with this method.

Usage

BayesianDeepLearning$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

if (FALSE) { # \dontrun{
# Initialize BDL module
bdl <- BayesianDeepLearning$new()

# Fit model with spatial data
bdl$fit_model(field_data, existing_samples)

# Predict with uncertainty
predictions <- bdl$predict_with_uncertainty(new_locations)

# Calculate specific uncertainty types
epistemic <- bdl$epistemic_uncertainty(predictions)
} # }


## ------------------------------------------------
## Method `BayesianDeepLearning$new`
## ------------------------------------------------

if (FALSE) { # \dontrun{
bdl <- BayesianDeepLearning$new()
bdl_custom <- BayesianDeepLearning$new(config = list(dropout_rate = 0.2))
} # }


## ------------------------------------------------
## Method `BayesianDeepLearning$fit_model`
## ------------------------------------------------

if (FALSE) { # \dontrun{
model <- bdl$fit_model(
  field_data = field_data,
  existing_samples = samples,
  target_variable = "soil_property",
  epochs = 100
)
} # }


## ------------------------------------------------
## Method `BayesianDeepLearning$predict_with_uncertainty`
## ------------------------------------------------

if (FALSE) { # \dontrun{
predictions <- bdl$predict_with_uncertainty(
  locations = new_locations,
  n_samples = 100
)
} # }


## ------------------------------------------------
## Method `BayesianDeepLearning$mc_dropout_predict`
## ------------------------------------------------

if (FALSE) { # \dontrun{
mc_predictions <- bdl$mc_dropout_predict(
  model = fitted_model,
  data = prediction_data,
  n_iterations = 100,
  dropout_rate = 0.1
)
} # }


## ------------------------------------------------
## Method `BayesianDeepLearning$variational_inference`
## ------------------------------------------------

if (FALSE) { # \dontrun{
vi_result <- bdl$variational_inference(
  data = training_data,
  prior_params = list(mean = 0, std = 1),
  n_samples = 1000
)
} # }


## ------------------------------------------------
## Method `BayesianDeepLearning$epistemic_uncertainty`
## ------------------------------------------------

if (FALSE) { # \dontrun{
epistemic <- bdl$epistemic_uncertainty(predictions)
} # }


## ------------------------------------------------
## Method `BayesianDeepLearning$aleatoric_uncertainty`
## ------------------------------------------------

if (FALSE) { # \dontrun{
aleatoric <- bdl$aleatoric_uncertainty(predictions)
} # }


## ------------------------------------------------
## Method `BayesianDeepLearning$total_uncertainty`
## ------------------------------------------------

if (FALSE) { # \dontrun{
total <- bdl$total_uncertainty(predictions)
} # }