DeepGLM class
Create DeepGLM model object
Syntax
Mdl = DeepGLM(Network,Name,Value)
Description
Mdl = DeepGLM(Network,Name,Value)
returns DeepGLM model object Mdl
given neural network structure Network
. Name,Value
specifies additional options using one or more name-value pair arguments. For example, users can specify the activation function or distribution of the output.
See: Input Arguments, Output Argument, Examples
Input Arguments
Data type: Array of positive integer
Neuron Network structure of the deepGLM model. [NumFeatures, L1,...,LM]
:
Numfeatures
- Number of features (columns) of training data.
L1,...,LM
- Number of hidden nodes in each hidden layer. For example,
L1
is the number of hidden nodes in the first hidden layer and so on.
Note: The output layer has only 1 node.
Name-Value Pair Arguments
Specify optional comma-separated pairs of Name,Value
arguments. Name
is the argument name and Value
is the corresponding value. Name
must appear inside quotes. You can specify several name and value pair arguments in any order as Name1,Value1,...,NameN,ValueN
.
Example: 'Distribution','Normal','Activation','relu'
specifies that the distribution of the response is normal, and the activation function of hidden layers is the Rectified Linear Unit (ReLU) function.
'Distribution' - Distribution of the response variable
Data Type: String
Distribution of the response variable, specified as the comma-separated pair consisting of 'Distribution' and one of the following:
'Normal' | Normal distribution (default) |
'Binomial' | Binomial distribution |
'Poisson' | Poisson distribution |
Example: 'Distribution','Normal'
'Activation' - Activation function of hidden layers
Data Type: String
Activation function of hidden layers, specified as the comma-separated pair consisting of 'Activation' and one of the following:
'relu' | $f(x)= \text{max}(x,0)$ | Rectified Linear Unit (ReLU) function (default) |
'sigmoid' | $f(x)=\frac{1}{1+e^{-x}}$ | Sigmoid function |
'tanh' | $f(x) = \frac{e^{2x}-1}{e^{2x}+1}$ | Tanh function |
Example: 'Activation','relu'
Output Arguments
Data type: DeepGLM Object
DeepGLM is an object of the DeepGLM class with pre-defined properties and functions.
Object Properties
The DeepGLM object properties include information about model-specific information, coefficient estimates and fitting method.
Properties | Data type | Description{: .text-center} |
---|---|---|
ModelName | string (r) | Name of the model, which is 'DeepGLM' |
NumParams | integer (+) | Number of model parameters |
Network | array | Neural network structure of DeepGLM models |
Distribution | string | Neural network structure of DeepGLM models |
Activation | string | Neural network structure of DeepGLM models |
Post * | struct | • Information about the fittedd method used to estimate model paramters • The DeepGLM model can only be fitted by NAGVAC and VAFC techniques |
Coefficient * | cell array | • Estimated Mean of weights of Deep Neural Network • Used to doing point estimation for new test data |
CoefficientVar * | cell array (r) | Variance of coefficient estimates |
Shrinkage * | array | Array storing estimated values of group Lasso coefficients |
LogLikelihood * | double (r) | Loglikelihood of the fitted model. |
FittedValue * | array (r) | • Fitted (predicted) values based on the input data. • For binary response, these are fitted probability |
Notation:
- * $\rightarrow$ object properties which are only available when the model is fitted. Default value is
None
. - (+) $\rightarrow$ positive number.
- (r) $\rightarrow$ read-only properties.
Object Methods
Use the object methods to initialize model parameters, predict responses, and to visualize the prediction.
vbayesInit | Initialization method of model parameters |
vbayesPredict | Predict responses of fitted DeepGLM models |
Examples
To be updated…
Reference
[1] Tran, M.-N., Nguyen, T.-N., Nott, D., and Kohn, R. (2020). Bayesian deep net GLM and GLMM. Journal of Computational and Graphical Statistics, 29(1):97-113. Read the paper
See Also
LogisticRegression $\mid$ RECH $\mid$ Custom model $\mid$ NAGVAC $\mid$ VAFC