These changelogs reflect the history of the Kili Technology service.
June 2, 2020
- Support: To ease support & provide more transparency on tickets, you can now access our support here and follow you requests. All requests are to be sent to firstname.lastname@example.org.
- APP / Review: It is now possible to search on the exact content of a label: "response:..."
- Kili Playground: Active learning is one of the most powerful lever to reduce the amount of labeled data requested to train a model. We released a new Jupyter notebook acting as a recipe to help the community leverage active learning.
- Kili Playground: How to make a NLP sentiment analysis or chatbot utterances / intent variations classification model robust ? Counterfactual data augmentaion is one of the key lever to reach this goal. We released a new Jupyter notebook acting as a recipe to show how to leverage this approach in Kili.
- APP / Performance: When you label, switching from an asset to another is faster.
- APP / Security: Overall security of the application has been improved: remove support of SSL1, implementation of HSTS.
- APP / IMAGE: Content of nested transcription fields is managed as expected.
- APP / IMAGE: On review, points are well displayed, and the color is centered.
- APP / NER: When you label a very long string, it is well managed in the job renderer.
- APP / Review: You can easily again share the URL of a labeled asset with other project members.
May 22, 2020
- Kili Playground: DICOM standard* is a well-known format that facilitates interoperability between medical imaging systems from different vendors. We wrote a Jupyter notebook acting as a recipe to manipulate DICOM data within Kili.
- APP: We worked on improve the scalibility of the overall infrastructure guarantees a faster UX on the application.
- APP: Labelers had no access to their KPIs. They can now access their own KPIs on the tab Audit labelers
- Python API: Exploit GraphQL's flexibility directly from Python. GraphQL offers the great flexibility to query any fields you want. We wanted to integrate this feature to our Python SDK. From now on, you can request particular fields by explicitly provide a list of field (
fields) to all query resolver (
assets, etc). This feature will prove particularly useful when you only want to speed up your queries and query only one or several particular fields (for instance,
id). The usage is documented in this recipes.
- APP: in the project list it is now possible to search with the project ID
- APP: autosave has been improved to help save labeler work every 30s
- DOC: Kili's Python documentation is now much more readable. Find the link here.
- App: JSON interface: when I create a new category, I see it has a property "children: " in JSON
- API: the attribution of specific assets to specific labeler
toBeLabelledBycan be unset by
update_properties_in_assetwith an empty list.
- APP: OCR project with pre ocerised documents has been migrated on the new interface
May 5, 2020
- APP : We now allow you to add custom colours to better suit your needs in
jsonInterface. See here
- APP: When you edit your interface using the JSON, a sanity check is now performed to prevent corrupted interfaces.
- APP: On classification task, when you use the dropdown component, you can choose to have it mandatory or not.
- DOC : You have the full list of GraphQL resolvers in documentation.
- DOC : You have the full list of Python resolvers in documentation
- API - Standardized new resolvers
labelsallow you to query all assets/projects/organizations/labels in one standardized manner.
- APP: Performance on the application on classification task has been improved.
- API: Warning: The following methods are deprecated since: 30/04/2020. The following methods will be removed after: 30/05/2020. See details.
- APP: When you drag & drop more than 500 assets in a row, you see the advice to use the API.
- APP: When you delete a user from the organisation panel or settings panel, you see it deleted.
- APP / NER: When you change the class of an entity using the shortcuts, you can do it in one click.
April 21, 2020
- APP / TEXT : a click on an object on the text unfolds the object in the job panel.
- APP / PDF: we just launched a BETA version of classification & named entities recognition for PDF. See here
- APP / NER: you can now configure more than one job for named entity recognition.
- APP / ALL INTERFACES: labels are autosaved on labeling interface. Whenever you create a new annotation, it is automatically saved, so you never lose your work.
- Python API: New resolvers to query
projectUsersaccording to several filters like their organizations, their names, their emails, their projects, etc.
- A complete recipe shows in details data formats of both
jsonInterface(parameter of labeling interface) and
jsonResponse(parameter of labels). As other recipes, this recipe is fully tested on Travis.
- DOC: Kili's documentation is now located at cloud.kili-technology.com/docs.
- APP: locks to manage the queue of asset has been improve to be more robust.
- Python API: we migrated the field
jsonInterface. All corresponding resolvers are flagged as deprecated in
- Python API: we removed all warnings when authenticating.
- APP / Review: switch between a skipped label and a default label is well managed
April 15, 2020
- APP / IMAGE & TEXT : an arrow, at instructions level, allows to deploy all objects of the job in one click.
- APP / IMAGE & TEXT : a coloured sticker at class level, help to identify objects on the viewer.
- APP / IMAGE : a click on an object on the image unfolds the object in the job panel.
- APP: a simplification of the data model for nested job has been implemented. It improves readability.
April 8, 2020
- APP / TEXT / NER / Relations: it is now possible to build multi destination terms relations: one entity as a start and several entities as end relations. See here.
- In Audit labeler, the number of annotations is now the number of default annotation. It doesn’t include reviews and help to have a better understanding of the contribution of each labeler.
- APP: the possibility to display meta data on all interfaces has been migrated, and is available on all interfaces. See here in the documentation.
- DOC: documentation has been simplified and readability improved. See doc.
- Security: improvements to better manage roles across the application and the API.
- APP: it is now possible to delete project with a corrupted JSON settings.
- APP / TEXT: text documents containing extension .docx for instance, are now correctly managed.
April 1, 2020
- APP: the deletion of a project is now available in the interface. In Settings / Admin. See here in the documentation.
- APP: interface render width can now be customised to improved readability for complexes tasks or transcription tasks for instance. In Settings / Edit you interface / JSON. See here in the documentation.
- APP: dropdown classification can now have a nested classification task.
- APP / TEXT / NER: the possibility to build relations has been migrated in the new interface builder. In Settings / Edit your interface. See here in the documentation.
- APP / IMAGE / Segmentation: the segmentation tool has been migrated in the new interface builder. In Settings / Edit your interface See here in the documentation.
- APP: the JSON for interface configuration is now editable as text. It simplifies modification. See in Settings / Edit your interface / JSON. See here in the documentation.
- APP / IMAGE / POLYLINE: when I draw a polyline outside the image, I see points are shrinked up to the border of the image.
- APP / Nested task: when a nested task is required, it remains greyed out until completion.
- APP / Interface builder: when I move a task under another task, I can save my modifications.