Hey @MichaelHauser99!
The best way to leverage video data in DataRobot is by extracting key frames (as images) of the video for each observation/row, and then using DataRobot's existing Visual AI functionality to create models with those images. Depending on the use case, you may want to extract an image every X seconds, or you may want to extract pre-tagged frames of key events in the video.
When building out the training dataset for such a video use case, the tabular modeling data will need to contain a column/feature for each image extracted from each video, and that feature should contain either (1) the relative file path to that image in a .zip directory, or (2) the base64 encoding of the image itself (as text). In an example where you take a frame every 0.5 seconds, you might label these columns something like "frame_0.0", "frame_0.5", "frame_1.0", etc. Note that there is currently no way to ingest raw video data directly into DataRobot for modeling.
For reference, here is some documentation further explaining how to upload images to the AI catalog or directly to a DataRobot project for modeling. Let us know what follow-up questions you have!
日本語
