OCI에서는 텍스트 또는 이미지로 구성된 데이터셋을 AI 또는 ML 에서 활용할 수 있도록 라벨링 할 수 있는 서비스인 Data Labeling 서비스를 제공합니다.
기존 V2 까지 공개되었던 데이터 라벨링 서비스에서 이번 업데이트를 통해 V3를 공개하였습니다. V3에서 추가된 기능은 아래와 같습니다.
CLASSIFICATION(분류)) 의 경우 FIRST_LETTER_MATCH(첫글자 매칭), FIRST_REGEX_MATCH(정규식 매칭), CUSTOM_LABELS_MATCH(사용자 정의 매칭) 유형을 제공함.
- FIRST_LETTER_MATCH: The first letter of the DLS Dataset record's name must be equal to the first letter of the label that the record will be annotated with. The matching is not case-sensitive.
Consider a dataset having following records: cat1.jpeg, cat2.jpeg, dog1.png, dog2.png
Label Set: cat , dog
Result of FIRST_LETTER_MATCH labeling algorithm:
cat1.jpeg will be labeled with cat label
cat2.jpeg will be labeled with cat label
dog1.png will be labeled with dog label
dog2.png will be labeled with dog label
- FIRST_REGEX_MATCH: The regular expression (regex) pattern assigned to _FIRST_MATCH_REGEX_PATTERN_ will be applied to the DLS Dataset record's name, and the first capture group extracted must be equal to the label that the record will be annotated with.
Consider a dataset having following records: PetCat1.jpeg, PetCat2.jpeg, PetDog1.png, PetDog2.png
Label Set: cat , dog
FIRST_MATCH_REGEX_PATTERN : ^([^\/]*)\/.*$
Result of FIRST_REGEX_MATCH labeling algorithm:
PetCat1.jpeg will be labeled with cat label
PetCat2.jpeg will be labeled with cat label
PetDog1.png will be labeled with dog label
PetDog2.png will be labeled with dog label
- CUSTOM_LABELS_MATCH: This algorithm takes object storage path as input along with the label that needs to be applied to records under that path. Only root level path is supported. Multiple labels can also be assigned to a given path. The labeling algorithm for this case is .
Consider a dataset having following records:
cat/cat1.jpeg, cat/cat2.jpeg, dog/dog1.jpeg, dog/dog2.jpeg
Labels in dataset: dog, pup, cat, kitten
CUSTOM_LABELS={ "dog/": ["dog","pup"], "cat/": ["cat", "kitten"] }
Result of CUSTOM_LABELS_MATCH algorithm:
cat/cat1.jpeg will be labeled with cat and kitten labels
cat/cat2.jpeg will be labeled with cat and kitten labels
dog/dog1.png will be labeled with dog and pup labels
dog/dog2.png will be labeled with dog and pup labels
OCI 데이터 과학의 Jupyter Notebook 세션에 Notebook Explorer 기능이 추가되었습니다. 이 기능을 통해서 다양한 콘다환경의 Notebook Example을 콘다환경 설치 없이 탐색하고 확인할 수 있게 되었습니다.
Young Hwan RELEASE-NOTES-2022-AIML
oci-release-notes-2022 june-2022 AI/ML