i.TSSKrSSKrSSKJr SSKJr SSKJrJ r SSK J r SSK J r SSKJr SS KJr SS KJr SSKr\R*R-S 5(a\eSSKJr \R4"\R6S 5r"S S5r"SS5rg!\a SSKrN>f=f)N)JSONDecodeError)ApiError)AsyncClientWrapperSyncClientWrapper)jsonable_encoder)remove_none_from_dict)UnprocessableEntityError)ClassifyResponse)HttpValidationErrorz1..c\rSrSrS\4SjrSS\\\\S.S\R\ S\R\ S\ S \R\R\ S \R\ S \R\ S \R\ S \ 4SjjrSrg)ClassifierClientclient_wrappercXlgN_client_wrapperselfrs a/home/james-whalen/.local/lib/python3.13/site-packages/llama_cloud/resources/classifier/client.py__init__ClassifierClient.__init__-N project_idorganization_idfilesfile_idsmatching_thresholdenable_metadata_heuristicrr rules_jsonrr r!r"returnc 8SU0nU[LaXHS'U[LaXXS'U[LaXhS'U[LaXxS'URRRS[R R URR5S3S5[XS .5[U5URR5S S 9n S U Rs=::aS :a,O O)[R"[U R55$U RS:Xa2[![R"["U R555eU R5n ['U RU S9e![$a ['U RU R(S9ef=f)x **[BETA]** Classify documents based on provided rules - simplified classification system. **This is a Beta feature** - API may change based on user feedback. This endpoint supports: - Classifying new uploaded files - Classifying existing files by ID - Both new files and existing file IDs in one request ## v0 Features: - **Simplified Rules**: Only `type` and `description` fields needed - **Matching Threshold**: Confidence-based classification with configurable threshold - **Smart Classification**: Filename heuristics + LLM content analysis - **Document Type Filtering**: Automatically filters out non-document file types - **Fast Processing**: Uses LlamaParse fast mode + GPT-4.1-nano - **Optimized Performance**: Parses each file only once for all rules ## Simplified Scoring Logic: 1. **Evaluate All Rules**: Compare document against all classification rules 2. **Best Match Selection**: Return the highest scoring rule above matching_threshold 3. **Unknown Classification**: Return as "unknown" if no rules score above threshold This ensures optimal classification by: - Finding the best possible match among all rules - Avoiding false positives with confidence thresholds - Maximizing performance with single-pass file parsing ## Rule Format: ```json [ { "type": "invoice", "description": "contains invoice number, line items, and total amount" }, { "type": "receipt", "description": "purchase receipt with transaction details and payment info" } ] ``` ## Classification Process: 1. **Metadata Heuristics** (configurable via API): - **Document Type Filter**: Only process document file types (PDF, DOC, DOCX, RTF, TXT, ODT, Pages, HTML, XML, Markdown) - **Filename Heuristics**: Check if rule type appears in filename - **Content Analysis**: Parse document content once and use LLM for semantic matching against all rules 2. **Result**: Returns type, confidence score, and matched rule information ## API Parameters: - `matching_threshold` (0.1-0.99, default: 0.6): Minimum confidence threshold for acceptable matches - `enable_metadata_heuristic` (boolean, default: true): Enable metadata-based features ## Supported Document Types: **Text Documents**: pdf, doc, docx, rtf, txt, odt, pages **Web Documents**: html, htm, xml **Markup**: md, markdown ## Limits (Beta): - Maximum 100 files per request - Maximum 10 rules per request - Rule descriptions: 10-500 characters - Document types: 1-50 characters (alphanumeric, hyphens, underscores) **Beta Notice**: This API is subject to change. Please provide feedback! Parameters: - project_id: typing.Optional[str]. - organization_id: typing.Optional[str]. - rules_json: str. JSON string containing classifier rules - files: typing.Optional[typing.List[str]]. - file_ids: typing.Optional[str]. - matching_threshold: typing.Optional[float]. - enable_metadata_heuristic: typing.Optional[bool]. r#rr r!r"POST/api/v1/classifier/classifyrr<paramsjsonheaderstimeout, status_codebodyOMITr httpx_clientrequesturllibparseurljoin get_base_urlr r get_headersr5pydantic parse_obj_asr r.r r rrtext rrrr#rr r!r"_request _response_response_jsons rclassify_documents#ClassifierClient.classify_documentss{J3? 1K   %W  4 #+Z T )-?) * $D 04M0 1((55==  LL D$8$8$E$E$G#H!JLh i( )gh!(+((446 >  )'' -# -(()99>>;KL L  C '*8+@+@ATV_VdVdVf+gh h S&^^-N9#8#8~NN Sy'<'<9>>R R Ss E00)Fr)__name__ __module__ __qualname____firstlineno__rrr8typingOptionalstrListfloatboolr rG__static_attributes__rrrrs.*;. ,00437)-59;?~OOOC(~O - ~O  ~O v{{3/0 ~O//#&~O#OOE2~O$*??4#8~O ~O~Orrc\rSrSrS\4SjrSS\\\\S.S\R\ S\R\ S\ S \R\R\ S \R\ S \R\ S \R\ S \ 4SjjrSrg)AsyncClassifierClientrcXlgrrrs rrAsyncClassifierClient.__init__rrNrrrr#rr r!r"r$c T# SU0nU[LaXHS'U[LaXXS'U[LaXhS'U[LaXxS'URRRS[R R URR5S3S5[XS .5[U5URR5S S 9IS hvN n S U Rs=::aS:a,O O)[R"[U R55$U RS:Xa2[![R"["U R555eU R5n ['U RU S9eN![$a ['U RU R(S9ef=f7f)r&r#rr r!r"r'r(r)r*r+r,Nr1r2r3r4r7rCs rrG(AsyncClassifierClient.classify_documentssJ3? 1K   %W  4 #+Z T )-?) * $D 04M0 1..;;CC  LL D$8$8$E$E$G#H!JLh i( )gh!(+((446 D   )'' -# -(()99>>;KL L  C '*8+@+@ATV_VdVdVf+gh h S&^^-N9#8#8~NN!  Sy'<'<9>>R R Ss+C F( E: B F(E<&F(<)F%%F(r)rIrJrKrLrrr8rMrNrOrPrQrRr rGrSrTrrrVrVs.*<. ,00437)-59;?~OOOC(~O - ~O  ~O v{{3/0 ~O//#&~O#OOE2~O$*??4#8~O ~O~OrrV)rM urllib.parser; json.decoderrcore.api_errorrcore.client_wrapperrrcore.jsonable_encoderrcore.remove_none_from_dictr !errors.unprocessable_entity_errorr types.classify_responser types.http_validation_errorr r@ __version__ startswith ImportError pydantic.v1v1castAnyr8rrVrTrrrls(&H5?I7>&&t,," {{6::s#BOBOJBOBOYs,B B'&B'