Operators are a collection of basic processes that assist in data modification, cleaning, filtering, deduplication, etc. We support a wide range of data sources and file formats, and allow for flexible extension to custom datasets.
This page offers a basic description of the operators (OPs) in Data-Juicer. Users can refer to the API documentation for the specific parameters of each operator. Users can refer to and run the unit tests (tests/ops/...) for examples of operator-wise usage as well as the effects of each operator when applied to built-in test data samples.
The operators in Data-Juicer are categorized into 5 types.
| Type | Number | Description |
|---|---|---|
| Formatter | 9 | Discovers, loads, and canonicalizes source data |
| Mapper | 63 | Edits and transforms samples |
| Filter | 44 | Filters out low-quality samples |
| Deduplicator | 8 | Detects and removes duplicate samples |
| Selector | 4 | Selects top samples based on ranking |
| Grouper | 2 | Group samples to batched samples |
| Aggregator | 3 | Aggregate for batched samples, such as summary or conclusion |
All the specific operators are listed below, each featured with several capability tags.
- Domain Tags
: general purpose
: specific to LaTeX source files
: specific to programming codes
: closely related to financial sector
- Modality Tags
: specific to text
: specific to images
: specific to audios
: specific to videos
: specific to multimodal
- Language Tags
: English
: Chinese
- Resource Tags
: only requires CPU resource (default)
: requires GPU/CUDA resource as well
| Operator | Tags | Description | Source code | Unit tests |
|---|---|---|---|---|
| local_formatter |    |
Prepares datasets from local files | code | tests |
| remote_formatter | |
Prepares datasets from remote (e.g., HuggingFace) | code | tests |
| csv_formatter | |
Prepares local .csv files |
code | tests |
| tsv_formatter | |
Prepares local .tsv files |
code | tests |
| json_formatter | |
Prepares local .json, .jsonl, .jsonl.zst files |
code | - |
| parquet_formatter | |
Prepares local .parquet files |
code | tests |
| text_formatter | |
Prepares other local text files (complete list) | code | - |
| empty_formatter | |
Prepares an empty dataset | code | tests |
| mixture_formatter | |
Handles a mixture of all the supported local file types | code | tests |
| Operator | Tags | Description | Source code | Unit tests |
|---|---|---|---|---|
| audio_ffmpeg_wrapped_mapper |  |
Simple wrapper to run a FFmpeg audio filter | code | tests |
| calibrate_qa_mapper |  |
Calibrate question-answer pairs based on reference text | code | tests |
| calibrate_query_mapper | |
Calibrate query in question-answer pairs based on reference text | code | tests |
| calibrate_response_mapper | |
Calibrate response in question-answer pairs based on reference text | code | tests |
| chinese_convert_mapper | |
Converts Chinese between Traditional Chinese, Simplified Chinese and Japanese Kanji (by opencc) | code | tests |
| clean_copyright_mapper |  |
Removes copyright notice at the beginning of code files (must contain the word copyright) | code | tests |
| clean_email_mapper | |
Removes email information | code | tests |
| clean_html_mapper | |
Removes HTML tags and returns plain text of all the nodes | code | tests |
| clean_ip_mapper | |
Removes IP addresses | code | tests |
| clean_links_mapper | |
Removes links, such as those starting with http or ftp | code | tests |
| expand_macro_mapper |  |
Expands macros usually defined at the top of TeX documents | code | tests |
| extract_entity_attribute_mapper | |
Extract attributes for given entities from the text. | code | tests |
| extract_entity_relation_mapper | |
Extract entities and relations in the text for knowledge graph. | code | tests |
| extract_event_mapper | |
Extract events and relevant characters in the text. | code | tests |
| extract_keyword_mapper | |
Generate keywords for the text. | code | tests |
| extract_nickname_mapper | |
Extract nickname relationship in the text. | code | tests |
| extract_support_text_mapper | |
Extract support sub text for a summary. | code | tests |
| fix_unicode_mapper | |
Fixes broken Unicodes (by ftfy) | code | tests |
| generate_qa_from_examples_mapper |  |
Generate question and answer pairs based on examples. | code | tests |
| generate_qa_from_text_mapper | |
Generate question and answer pairs from text. | code | tests |
| image_blur_mapper |  |
Blur images | code | tests |
| image_captioning_from_gpt4v_mapper |  |
generate samples whose texts are generated based on gpt-4-visison and the image | code | - |
| image_captioning_mapper | |
generate samples whose captions are generated based on another model (such as blip2) and the figure within the original sample | code | tests |
| image_diffusion_mapper | |
Generate and augment images by stable diffusion model | code | tests |
| image_face_blur_mapper | |
Blur faces detected in images | code | tests |
| image_tagging_mapper | |
Mapper to generate image tags from the input images. | code | tests |
| nlpaug_en_mapper | |
Simply augments texts in English based on the nlpaug library |
code | tests |
| nlpcda_zh_mapper | |
Simply augments texts in Chinese based on the nlpcda library |
code | tests |
| optimize_qa_mapper | |
Optimize both the query and response in question-answering samples. | code | tests |
| optimize_query_mapper | |
Optimize the query in question-answering samples. | code | tests |
| optimize_response_mapper | |
Optimize the response in question-answering samples. | code | tests |
| pair_preference_mapper | |
Construct paired preference samples. | code | tests |
| punctuation_normalization_mapper | |
Normalizes various Unicode punctuations to their ASCII equivalents | code | tests |
| python_file_mapper | |
Executing Python function defined in a file | code | tests |
| python_lambda_mapper | |
Executing Python lambda function on data samples | code | tests |
| relation_identity_mapper | |
Identify relation between two entity in the text. | code | tests |
| remove_bibliography_mapper | |
Removes the bibliography of TeX documents | code | tests |
| remove_comments_mapper | |
Removes the comments of TeX documents | code | tests |
| remove_header_mapper | |
Removes the running headers of TeX documents, e.g., titles, chapter or section numbers/names | code | tests |
| remove_long_words_mapper | |
Removes words with length outside the specified range | code | tests |
| remove_non_chinese_character_mapper | |
Remove non Chinese character in text samples. | code | tests |
| remove_repeat_sentences_mapper | |
Remove repeat sentences in text samples. | code | tests |
| remove_specific_chars_mapper | |
Removes any user-specified characters or substrings | code | tests |
| remove_table_text_mapper |  |
Detects and removes possible table contents (:warning: relies on regular expression matching and thus fragile) | code | tests |
| remove_words_with_incorrect_ substrings_mapper | |
Removes words containing specified substrings | code | tests |
| replace_content_mapper | |
Replace all content in the text that matches a specific regular expression pattern with a designated replacement string | code | tests |
| sentence_split_mapper | |
Splits and reorganizes sentences according to semantics | code | tests |
| text_chunk_mapper | |
Split input text to chunks. | code | tests |
| video_captioning_from_audio_mapper | |
Caption a video according to its audio streams based on Qwen-Audio model | code | tests |
| video_captioning_from_frames_mapper | |
generate samples whose captions are generated based on an image-to-text model and sampled video frames. Captions from different frames will be concatenated to a single string | code | tests |
| video_captioning_from_summarizer_mapper | |
Generate video captions by summarizing several kinds of generated texts (captions from video/audio/frames, tags from audio/frames, ...) | code | tests |
| video_captioning_from_video_mapper | |
generate samples whose captions are generated based on another model (video-blip) and sampled video frame within the original sample | code | tests |
| video_extract_frames_mapper | |
extract frames from video files according to specified methods | code | tests |
| video_face_blur_mapper |  |
Blur faces detected in videos | code | tests |
| video_ffmpeg_wrapped_mapper | |
Simple wrapper to run a FFmpeg video filter | code | tests |
| video_remove_watermark_mapper | |
Remove the watermarks in videos given regions | code | tests |
| video_resize_aspect_ratio_mapper | |
Resize video aspect ratio to a specified range | code | tests |
| video_resize_resolution_mapper | |
Map videos to ones with given resolution range | code | tests |
| video_split_by_duration_mapper | |
Mapper to split video by duration | code | tests |
| video_split_by_key_frame_mapper | |
Mapper to split video by key frame | code | tests |
| video_split_by_scene_mapper | |
Split videos into scene clips | code | tests |
| video_tagging_from_audio_mapper | |
Mapper to generate video tags from audio streams extracted from the video. | code | tests |
| video_tagging_from_frames_mapper | |
Mapper to generate video tags from frames extracted from the video. | code | tests |
| whitespace_normalization_mapper | |
Normalizes various Unicode whitespaces to the normal ASCII space (U+0020) | code | tests |
| Operator | Tags | Description | Source code | Unit tests |
|---|---|---|---|---|
| alphanumeric_filter | |
Keeps samples with alphanumeric ratio within the specified range | code | tests |
| audio_duration_filter | |
Keep data samples whose audios' durations are within a specified range | code | tests |
| audio_nmf_snr_filter | |
Keep data samples whose audios' Signal-to-Noise Ratios (SNRs, computed based on Non-Negative Matrix Factorization, NMF) are within a specified range | code | tests |
| audio_size_filter | |
Keep data samples whose audios' sizes are within a specified range | code | tests |
| average_line_length_filter | |
Keeps samples with average line length within the specified range | code | tests |
| character_repetition_filter | |
Keeps samples with char-level n-gram repetition ratio within the specified range | code | tests |
| flagged_words_filter | |
Keeps samples with flagged-word ratio below the specified threshold | code | tests |
| image_aesthetics_filter | |
Keeps samples containing images whose aesthetics scores are within the specified range | code | tests |
| image_aspect_ratio_filter | |
Keeps samples containing images with aspect ratios within the specified range | code | tests |
| image_face_count_filter | |
Keeps samples containing images with face counts within the specified range | code | tests |
| image_face_ratio_filter | |
Keeps samples containing images with face area ratios within the specified range | code | tests |
| image_nsfw_filter | |
Keeps samples containing images with NSFW scores below the threshold | code | tests |
| image_pair_similarity_filter | |
Keeps image pairs with image feature cosine similarity within the specified range based on a CLIP model | code | tests |
| image_shape_filter | |
Keeps samples containing images with widths and heights within the specified range | code | tests |
| image_size_filter | |
Keeps samples containing images whose size in bytes are within the specified range | code | tests |
| image_text_matching_filter | |
Keeps samples with image-text classification matching score within the specified range based on a BLIP model | code | tests |
| image_text_similarity_filter | |
Keeps samples with image-text feature cosine similarity within the specified range based on a CLIP model | code | tests |
| image_watermark_filter | |
Keeps samples containing images with predicted watermark probabilities below the threshold | code | tests |
| language_id_score_filter | |
Keeps samples of the specified language, judged by a predicted confidence score | code | tests |
| maximum_line_length_filter | |
Keeps samples with maximum line length within the specified range | code | tests |
| perplexity_filter | |
Keeps samples with perplexity score below the specified threshold | code | tests |
| phrase_grounding_recall_filter | |
Keeps samples whose locating recalls of phrases extracted from text in the images are within a specified range | code | tests |
| special_characters_filter | |
Keeps samples with special-char ratio within the specified range | code | tests |
| specified_field_filter | |
Filters samples based on field, with value lies in the specified targets | code | tests |
| specified_numeric_field_filter | |
Filters samples based on field, with value lies in the specified range (for numeric types) | code | tests |
| stopwords_filter | |
Keeps samples with stopword ratio above the specified threshold | code | tests |
| suffix_filter | |
Keeps samples with specified suffixes | code | tests |
| text_action_filter | |
Keeps samples containing action verbs in their texts | code | tests |
| text_entity_dependency_filter | |
Keeps samples containing dependency edges for an entity in the dependency tree of the texts | code | tests |
| text_length_filter | |
Keeps samples with total text length within the specified range | code | tests |
| token_num_filter | |
Keeps samples with token count within the specified range | code | tests |
| video_aesthetics_filter | |
Keeps samples whose specified frames have aesthetics scores within the specified range | code | tests |
| video_aspect_ratio_filter | |
Keeps samples containing videos with aspect ratios within the specified range | code | tests |
| video_duration_filter | |
Keep data samples whose videos' durations are within a specified range | code | tests |
| video_frames_text_similarity_filter | |
Keep data samples whose similarities between sampled video frame images and text are within a specific range | code | tests |
| video_motion_score_filter | |
Keep samples with video motion scores within a specific range | code | tests |
| video_motion_score_raft_filter | |
Keep samples with video motion scores (based on RAFT model) within a specific range | code | tests |
| video_nsfw_filter | |
Keeps samples containing videos with NSFW scores below the threshold | code | tests |
| video_ocr_area_ratio_filter | |
Keep data samples whose detected text area ratios for specified frames in the video are within a specified range | code | tests |
| video_resolution_filter | |
Keeps samples containing videos with horizontal and vertical resolutions within the specified range | code | tests |
| video_watermark_filter | |
Keeps samples containing videos with predicted watermark probabilities below the threshold | code | tests |
| video_tagging_from_frames_filter | |
Keep samples containing videos with given tags | code | tests |
| words_num_filter | |
Keeps samples with word count within the specified range | code | tests |
| word_repetition_filter | |
Keeps samples with word-level n-gram repetition ratio within the specified range | code | tests |
| Operator | Tags | Description | Source code | Unit tests |
|---|---|---|---|---|
| document_deduplicator | |
Deduplicates samples at document-level by comparing MD5 hash | code | tests |
| document_minhash_deduplicator | |
Deduplicates samples at document-level using MinHashLSH | code | tests |
| document_simhash_deduplicator | |
Deduplicates samples at document-level using SimHash | code | tests |
| image_deduplicator | |
Deduplicates samples at document-level using exact matching of images between documents | code | tests |
| video_deduplicator | |
Deduplicates samples at document-level using exact matching of videos between documents | code | tests |
| ray_document_deduplicator | |
Deduplicates samples at document-level by comparing MD5 hash on ray | code | - |
| ray_image_deduplicator | |
Deduplicates samples at document-level using exact matching of images between documents on ray | code | - |
| ray_video_deduplicator | |
Deduplicates samples at document-level using exact matching of videos between documents on ray | code | - |
| Operator | Tags | Description | Source code | Unit tests |
|---|---|---|---|---|
| frequency_specified_field_selector | |
Selects top samples by comparing the frequency of the specified field | code | tests |
| random_selector | |
Selects samples randomly | code | tests |
| range_specified_field_selector | |
Selects samples within a specified range by comparing the values of the specified field | code | tests |
| topk_specified_field_selector | |
Selects top samples by comparing the values of the specified field | code | tests |
| Operator | Tags | Description | Source code | Unit tests |
|---|---|---|---|---|
| key_value_grouper | |
Group samples to batched samples according values in given keys. | code | tests |
| naive_grouper | |
Group all samples to one batched sample. | code | tests |
| Operator | Tags | Description | Source code | Unit tests |
|---|---|---|---|---|
| entity_attribute_aggregator | |
Return conclusion of the given entity's attribute from some docs. | code | tests |
| most_relavant_entities_aggregator | |
Extract entities closely related to a given entity from some texts, and sort them in descending order of importance. | code | tests |
| nested_aggregator | |
Considering the limitation of input length, nested aggregate contents for each given number of samples. | code | tests |
We welcome contributions of adding new operators. Please refer to How-to Guide for Developers.