Skip to main content
Version: Next

Demo (preprocessed) datasets

Overview​

In H2O Hydrogen Torch, you can access demo (preprocessed) datasets to explore supported problem types.

Import a demo (preprocessed) dataset​

To import a demo (preprocessed) dataset to H2O Hydrogen Torch, consider the following instructions:

  1. In the H2O Hydrogen Torch navigation menu, click Import dataset.
  2. In the Source box, select AWS S3.
  3. In the S3 bucket name box, enter h2o-release/hydrogen-torch/1.4.0.
  4. In the File name list, select one of the Demo (preprocessed) datasets.
  5. Click Continue.
  6. Again, click Continue.
  7. Again, click Continue.
Note
  • After importing a preprocessed dataset, you can use it for an experiment.
  • To learn how to preprocess your dataset for a particular supported problem type, see Dataset formats

Demo (preprocessed) datasets​

Below are the demo (preprocessed) datasets in H2O Hydrogen Torch.

Image​

flower_image_classification.zip​

  • Description: This preprocessed dataset contains images of dandelions, daisies, roses, tulips, and sunflowers.
  • Dataset columns: image, label
  • Problem type: Image classification
note

To learn more about this dataset, see Flowers Dataset.

coins_image_regression.zip​

  • Description: This preprocessed dataset contains a collection of images with one or more coins. Each image has been labeled to indicate the sum of its coins. The currency of the coins is the Brazilian Real (R$).
  • Dataset columns: image_path, label, fold
  • Problem type: Image regression
note

To learn more about this dataset, see Brazilian Coins.

global_wheat_image_object_detection.zip​

  • Description: This preprocessed dataset consists of wheat field images annotated with bounding boxes around identified wheat heads.
  • Dataset columns: image, class_id, x_min, y_min, x_max, y_max
  • Problem type: Single-class object detection
note

To learn more about this dataset, see Global Wheat Dataset.

bicycle_image_metric_learning.zip​

  • Description: This preprocessed dataset contains images of online bicycle ads. Each ad has multiple images marked by their class ID.
  • Dataset columns: image, label, fold
  • Problem type: Image metric learning
note

To learn more about this dataset, see The Stanford Online Products dataset.

fashion_image_semantic_segmentation.zip​

  • Description: This preprocessed dataset contains images corresponding to fashion/apparel segmentations. As well, the dataset contains images of people wearing various clothing types in multiple poses.
  • Dataset columns: image, class_id, rle_mask
  • Problem type: Semantic segmentation
note

To learn more about this dataset, see Clothing Co-Parsing Dataset.

coco_image_instance_segmentation.zip​

  • Description: This preprocessed dataset contains a subsample of the famous Common Objects in Context (COCO) dataset. This subsample includes only a single "Car" class. In other words, all images contain a car or multiple cars.
  • Dataset columns: image_id, class_id, rle_mask
  • Problem type: Image instance segmentation
note

To learn more about this dataset, see COCO Dataset.

covid_ct_image_semantic_segmentation_3d.zip​

  • Description: This preprocessed dataset contains a collection of 20 3D computed tomography (CT) images depicting the human chest and lungs.
  • Dataset columns: image, class_id, rle_mask
  • Problem type: 3D image semantic segmentation
note

To learn more about this dataset, see COVID-19 CT scans.

mnist_3d_image_regression_3d.zip​

  • Description: This preprocessed dataset contains 60,000 3D digital images of numbers ranging from 0 to 9. The dataset was constructed by extracting images from the MNIST database.
  • Dataset columns: image, label
  • Problem type: 3D image regression
note

To learn more about the MNIST database, see The MNIST database of handwritten digits.

mnist_3d_image_classification_3d.zip​

  • Description: This preprocessed dataset contains 60,000 3D digital images of numbers ranging from 0 to 9. The dataset was constructed by extracting images from the MNIST database.
  • Dataset columns: image, label
  • Problem type: 3D image classification
note

To learn more about the MNIST database, see The MNIST database of handwritten digits.

Text​

amazon_reviews_text_classification.csv​

  • Description: This preprocessed dataset contains a collection of reviews from Amazon. Each review (in text form) includes the title of the review and the review itself. The dataset has been labeled to indicate whether a review is positive or negative.
  • Dataset columns: text, label
  • Problem type: Text classification
note

To learn more about this dataset, see Amazon product data.

cnn_dailymail_text_sequence_to_sequence.zip​

  • Description: This preprocessed dataset contains human-generated abstract summaries from news stories published on the CNN and Daily Mail websites.
  • Dataset columns: text, summary, id
  • Problem type: Text sequence to sequence
note

To learn more about this dataset, see abisee/cnn-dailymail.

wellformed_query_text_regression.csv​

  • Description: This preprocessed dataset contains a collection of search queries. Every query was rated between 0 and 1 specifying whether or not the query was well-formed.
  • Dataset columns: text, rating
  • Problem type: Text regression
note

To learn more about this dataset, see Query-wellformedness Dataset.

conll2003_text_token_classification.zip​

  • Description: This preprocessed dataset contains a collection of text pieces that have their name entities specified. Name entities refer to abstract or physical objects such as a person, product, etc., that can be indicated with a proper name.
  • Dataset columns: id, text, pos_tags, chunk_tags, ner_tags
  • Problem type: Text token classification
note

To learn more about this dataset, see Language-Independent Named Entity Recognition (II).

squad_text_span_prediction.zip​

  • Description: This preprocessed dataset contains questions with answers and contexts that can be used to answer the questions.
  • Dataset columns: question, context, answer
  • Problem type: Text span prediction
note

To learn more about this dataset, see The Stanford Question Answering Dataset.

ubuntu_text_metric_learning.zip​

  • Description: This preprocessed dataset contains a preprocessed collection of questions from AskUbuntu.com. Questions are grouped in similar clusters (label).
  • Dataset columns: text, label, fold
  • Problem type: Text metric learning
note

Image and text​

food_101_imageandtext_classification.zip​

  • Description: This preprocessed dataset contains recipe titles and images of 5 types of salads: Beet, Caesar, Caprese, Greek, and Seaweed.
  • Dataset columns: image, text, label
  • Problem type: Image and text classification
note

To learn more about this dataset, see Food-101.

Audio​

esc10_audio_classification.zip​

  • Description: This preprocessed dataset contains 5-second-long recordings organized into ten classes (with 40 examples per class). Clips in this dataset have been manually extracted from public field recordings gathered by the Freesound.org project.
  • Dataset columns: filename, fold, label
  • Problem type: Audio classification
note

To learn more about this dataset, see ESC-50: Dataset for Environmental Sound Classification.

amnist_audio_regression.zip​

  • Description: This preprocessed dataset contains a collection of 30,000 audio samples of spoken digits (0-9) of sixty different speakers.
  • Dataset columns: audio, label, fold
  • Problem type: Audio regression
note

To learn more about this dataset, see Audio MNIST.

Speech​

minds14_US_speech_recognition.zip​

  • Description: This preprocessed dataset contains a collection of 558 speech samples (EN-US subset) related to phone banking. The dataset was manually re-annotated as the transcriptions from the original dataset were generated using automated speech models.
  • Dataset columns: file, transcript, duration
  • Problem type: Speech recognition
note

To learn more about this dataset, see MInDS-14.

Graph​

ogbn-mag_graph_node_classification.zip​

  • Description: This preprocessed dataset represents a heterogeneous network derived from a subset of the Microsoft Academic Graph (MAG). It contains four types of entities: papers (736,389 nodes), authors(1,134,649 nodes), institutions (8,740 nodes), and fields of study (59,965 nodes). There are four types of directed relationships between these entities: an author is "affiliated with" an institution, an author "writes" a paper, a paper "cites" another paper, and a paper "has a topic of" a field of study. Each paper is associated with a 128-dimensional word2vec feature vector that captures its content, while the other entity types do not have additional features. This dataset is structured to facilitate the analysis of complex interrelationships among academic entities.
  • Dataset columns: node_id, label
  • Problem type: Graph node classification
note

To learn more about this dataset, see Dataset ogbn-mag.

ogbn-proteins_graph_node_regression.zip​

  • Description: This preprocessed dataset represents an undirected and weighted graph that illustrates various biological relationships between proteins. In this graph, each node represents a protein, while the edges denote different types of associations between these proteins, such as physical interactions, co-expression, or evolutionary relationships (homology). Each edge is associated with 8-dimensional features, where each dimension reflects the confidence level of a specific association type, with values ranging from 0 to 1 (with higher values indicating greater confidence). The proteins in this dataset are derived from 8 distinct species.
  • Dataset columns: label0 ... label111, node_id
  • Problem type: Graph node regression
note

To learn more about this dataset, see Dataset ogbn-proteins.

Feedback