Supported problem types¶
H2O Hydrogen Torch supports an array of diverse problem types in computer vision, natural language, and audio, and each, in turn, is explained below.
Image regression¶
Image regression refers to assigning one or more continuous target labels to an input image.
This problem type supports multi-label experiments.
To learn about the available settings for this problem type, see Experiment settings: Image regresssion.
Image classification¶
Image classification refers to assigning one or more categorical target labels to an input image; this includes binary classification, multi-class classification, and multi-label classification.
This problem type supports binary, multi-class, and multi-label experiments.
To learn about the available settings for this problem type, see Experiment settings: Image classification.
Image metric learning¶
Image metric learning refers to establishing similarity or dissimilarity between images.
To learn about the available settings for this problem type, see Experiment settings: Image metric learning.
Image object detection¶
Image object detection refers to locating individual objects in an image by drawing bounding boxes around them.
This problem type supports multi-class experiments.
To learn about the available settings for this problem type, see Experiment settings: Image object detection.
Image semantic segmentation¶
Image semantic segmentation refers to associating each pixel of an image with a class label (such as phones, pencils, or roads).
This problem type supports multi-class experiments.
To learn about the available settings for this problem type, see Experiment settings: Image semantic segmentation.
Image instance segmentation¶
Image instance segmentation refers to locating individual objects in an image by drawing masks around them.
This problem type supports multi-class experiments.
To learn about the available settings for this problem type, see Experiment settings: Image instance segmentation.
Text regression¶
Text regression refers to assigning one or more continuous target labels to an input text.
This problem type supports multi-label experiments.
To learn about the available settings for this problem type, see Experiment settings: Text regression.
Text classification¶
Text classification refers to assigning one or more categorical target labels to an input text; this includes binary classification, multi-class classification, and multi-label classification.
This problem type supports binary, multi-class, and multi-label experiments.
To learn about the available settings for this problem type, see Experiment settings: Text classification.
Text sequence to sequence¶
Text sequence to sequence (Seq2seq) refers to the task of predicting an output sequence given an input sequence; in other words, Seq2seq turns one sequence into another (sequence transformation).
To learn about the available settings for this problem type, see Experiment settings: Text sequence to sequence.
Text span prediction¶
Text span prediction refers to finding a substring in a larger context. A typical problem is question-answering, were given a context and question, the task is to find the answer (substring) in the context.
To learn about the available settings for this problem type, see Experiment settings: Text span prediction.
Text token classification¶
Text token classification refers to assigning a label to all tokens in a piece of text in contrast to text classification, where the entire text is given a label.
This problem type supports binary and multi-class classification experiments.
To learn about the available settings for this problem type, see Experiment settings: Text token classification.
Text metric learning¶
Text metric learning refers to establishing similarity or dissimilarity between texts.
To learn about the available settings for this problem type, see Experiment settings: Text metric learning.
Audio regression¶
Audio regression refers to assigning one or more continuous target labels to input audio.
This problem type supports multi-label experiments.
To learn about the available settings for this problem type, see Experiment settings: Audio regression.
Audio classification¶
Audio classification refers to assigning one or more categorical target labels to input audio; this includes binary classification, multi-class classification, and multi-label classification.
This problem type supports binary, multi-class, and multi-label experiments.
To learn about the available settings for this problem type, see Experiment settings: Audio classification.
- Submit and view feedback for this page
- Send feedback about H2O Hydrogen Torch to cloud-feedback@h2o.ai