The basic interface to Hive’s models is referred to as a project. Multiple projects can reside in a Hive account on thehive.ai. Each project provides access to a specific Hive machine learning model via a user interface (UI) as well as a programmatic interface (API). Each piece of data submitted to a project for processing is referred to as a task.
Collaborator: Only allowed to read/write with honeypot and qualifier features (N/A to automated Hive APIs)
Read: Access to view API projects
Write: Read + the ability to upload and export tasks in the UI
Admin/Owner: Write + the ability to update billing information
Please reach out to our sales team at [email protected] to help you get set-up with an account and demo projects.
A project will offer one of two types of prediction endpoints: Model-Only or Model + Manual Review. Model-Only provides results from Hive's state-of-the-art machine learning models, while Model + Manual Review improves on this accuracy by leveraging our distributed workforce to confirm a small percentage of cases where the models are uncertain.
Model + Manual Review
Our sales team can help you figure out which solution is best suited for your use case.
Tasks can be submitted via UI as well as via API. Clicking on a project after you log in takes you to the project dashboard (as seen in the image below). In your dashboard, you can view statistics about your project, upload data, view previously uploaded data, as well as export your data. There are two ways in which you can upload more data to your project using the UI. Both can be accessed by clicking on 'Upload More Data.'
Local Media Direct Upload
CSV Upload for Hosted Media
Drag and drop upload of local media files.
Suited for small submissions of 10-100 tasks.
Submit a CSV with public or signed URLs to hosted media files.
Suited for medium submissions of 100 - 50,000 tasks.
You can upload several local media files by selecting 'Upload files directly from your computer.' Drag and drop the files you wish to upload and finally hit 'Submit File.' Depending on the project, you can submit images, videos, or audio files.
Acceptable formats for model processing includes:
- These files may process successfully, but may not playback in all browsers in the UI. These formats will not work with projects that use human labeling.
The available CSV fields are:
URL: This column will contain the public or signed URLs for the media files to be submitted. This column is required if no Text Data (see below) is provided.
Callback URL (optional): This column may contain a Callback URL to a server. Results will be posted to that specified URL.
Callback Metadata (optional): This column may contain stringified text metadata (i.e. JSON or XML), which will be retained and returned in the task’s results.
Task Priority (optional): This column is a priority from 0-9, where 9 is the highest priority, such that you can prioritize different submissions.
Text Data (optional): For text models, this column accepts the text string to run through the model. If Text Data is provided, the URL should be left empty.
You must map the columns of your CSV to the respective fields as seen below:
Hive's new Model Update Feature allows you to update your API in your account whenever there is a new and improved model ready for production. You can find this in the Project Settings page.
- please contact your sales rep or [email protected] if you need to move back to an older config
You can view the results of submitted tasks directly in the UI by selecting 'View data' in the project dashboard. You then hover over the desired task and select 'View Task Details'. On this page you will find details about your task (i.e. Task ID, Callback Metadata, Completed On, etc.). To view the results of the task, you can then click on the 'Results' tab. Here you will find the predictions made on the task.
Exporting allows you to generate a CSV containing the results for your tasks. An export request accepts parameters to filter which tasks should be exported. After the export is ready, you may then download it.
Relevant filters include:
- Date Range: Choose all tasks submitted between two datetimes.
- Bulk Upload ID: Only select tasks submitted within a single upload (limited to media files or CSV; tasks submitted via API cannot be found this way). The bulk upload ID can be found in the Upload History tab on the project dashboard.
- Data Columns: Specifies which data fields to export.
a. Task ID: Universally unique identifier (UUID) per task.
b. Image URL: Originally uploaded URL.
c. Original Filename: Name of the locally uploaded file.
d. Status: JSON object with model results.
e. Timestamp (video and audio): Frame timestamp in seconds if the media was split into frames*.
f. Callback Metadata: Any metadata submitted by the user in the Callback Metadata field.
*Depending on the use case, some projects provide exports where each frame is split as an individual row in the CSV, while other projects will provide exports where a single row contains the results for all the frames of a media file.
Updated 4 days ago