Skip to main content

introduce

Intro​

Datumo Scope is a visualization-based dataset analysis solution.
Implement Data-centric AI by visually analyzing the coverage and edge cases of your training data.







Data distribution visualization​

untitled
Datumo Scope provides a flat graph that places similar data close together and dissimilar data far apart, allowing you to quickly understand the coverage of the entire dataset with easy manipulation and highly readable UI design.

*Feature space is a space expressed by compressing the feature vector, which consists of data information, into multi-dimensional points.

Datumo Scope automatically generates feature vectors or visualizes feature spaces using existing feature vectors.







Reflect metadata and model metrics​

untitled
Datumo Scope provides a data distribution graph that reflects metadata and model metric information. You can filter data in various ways by querying information such as data collection environments and model performance indicators.

*Different data collected under various weather and time conditions are separated and represented in different colors according to model performance metrics, making it easy to quickly identify edge cases.







Data curation​

untitled
Automatically selects data while minimizing damage to dataset coverage. You can freely set the number or ratio of data to be selected, as well as the selection algorithm. By using the curation function locally, you can create a more precise dataset.

*The curation function is used in various ways during the machine learning (ML) lifecycle, such as quickly analyzing the entire dataset coverage or classifying the dataset according to its purpose (Train/Test set split).







untitled
Automatically searches for data similar to the specified data. You can repeatedly search and edit the dataset within the specified range to configure the dataset as desired.

*You can select auxiliary data to adjust the search results to reflect your insights. You can also manually search for similar data by clicking on the surrounding points, but using the search function allows for more precise and efficient work.