"NeuralNet" (Resource Object Type)


  • Neural net resources contain neural networks in content elements that can be accessed with NetModel.


  • There are standard ResourceObject properties common to all resource types ». Additionally, each resource type defines additional special properties.
  • Special properties for neural net resources associated with the content include:
  • "ContentElements"list of content element names available via NetModel["name","elem"]
    "ContentElementLocations"storage locations of content elements
    "DefaultContentElement"name of content element available via NetModel["name"]
    "Format"formats of the content elements
    "ParameterizationData"for parameterized net models, stores internally used information
  • Special properties associated with the resource metadata include:
  • "TrainingSetData"link to training data
    "TrainingSetInformation"description of the data used to train the net
  • The "ContentElementLocations" property is an Association with content element names for keys and locations for values. Each value can be a CloudObject, LocalObject, File or URL.
  • Properties used for sorting data resources include:
  • "InputDomains"list input types ("Image", "Text", etc.)
    "TaskType"type of task performed by the net ("Classification","Regression")
  • All neural net resources have the property "ResourceType""NeuralNet".
  • Commonly used standard ResourceObject properties for data resources include:
  • "ExampleNotebook"notebook of example inputs and outputs
  • The "SourceMetadata" value is an Association that can include the following keys:
  • "Citation"source/reference citation
    "Creator"name of the author or creator
    "Date"original publication date
    "Rights"rights for the source
    "Source"link to the original source

Using a NeuralNet Resource

  • The nets within a ResourceObject are accessed with NetModel.
  • Properties can be accessed using ResourceObject[]["prop"].
  • Often, neural net resources have a notebook demonstrating the construction process available via NetModel["name","ConstructionNotebook"].
  • Neural net resources can include a single trained net or select from a parameterized collection of nets based on additional inputs to NetModel.


open allclose all

Basic Examples  (1)

Retrieve a neural net resource from the public repository:

The resource has type "NeuralNet":

Retrieve the default neural net:

Retrieve an input and use the net:

Scope  (2)

Explore the metadata for a neural net resource with one trained net:

See the names of the content elements:

See the locations and formats of the data files:

Open the example notebook:

See the task type:

See the input domains:

Read descriptions of the net and the training set:

Explore the metadata for a parameterized neural net resource with multiple trained nets available:

See the parameterization information:

See the names of the content elements. Note multiple evaluation nets:

Specify parameters to retrieve a net:

Retrieve the default net for comparison:

For the same input, the two parameterizations give different probabilities: