t-SNE Application Function

This function re-applies a t-SNE model that was trained with a tsne.train procedure type to new data points.

It is especially useful where a known layout has been produced, but some of the underlying data behind that layout has changed and a new layout that maps points in the same space is needed.

Input and Output Values

The t-SNE function takes the same input value as the procedures and outputs a single value named embedding with as many dimensions as the output was trained on.

Configuration

A new function of type tsne.embedRow named <id> can be created as follows:

mldb.put("/v1/functions/"+<id>, {
    "type": "tsne.embedRow",
    "params": {
        "modelFileUrl": <Url>
    }
})

with the following key-value definitions for params:

See also

• Wikipedia t-SNE Article
• The svd.embedRow function type is often used before the t-SNE function to turn a sparse dataset into an embedding.
• The tsne.train procedure type trains the model