tracebloc 1.2.85__tar.gz → 1.2.87__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {tracebloc-1.2.85 → tracebloc-1.2.87}/PKG-INFO +1 -1
- {tracebloc-1.2.85 → tracebloc-1.2.87}/pyproject.toml +1 -1
- tracebloc-1.2.87/tracebloc/raw_mode/__init__.py +91 -0
- tracebloc-1.2.87/tracebloc/raw_mode/classes.py +100 -0
- tracebloc-1.2.87/tracebloc/raw_mode/declaration.py +89 -0
- tracebloc-1.2.87/tracebloc/raw_mode/fitted_model.py +86 -0
- tracebloc-1.2.85/tracebloc/raw_mode_frame.py → tracebloc-1.2.87/tracebloc/raw_mode/frame_recording.py +4 -3
- tracebloc-1.2.87/tracebloc/raw_mode/link_check.py +58 -0
- tracebloc-1.2.87/tracebloc/raw_mode/pairs.py +54 -0
- tracebloc-1.2.87/tracebloc/raw_mode/schema.py +58 -0
- tracebloc-1.2.85/tracebloc/raw_mode_seed.py → tracebloc-1.2.87/tracebloc/raw_mode/seed_prefixes.py +1 -1
- tracebloc-1.2.87/tracebloc/raw_mode/sklearn_check.py +100 -0
- tracebloc-1.2.85/tracebloc/raw_mode_survival.py → tracebloc-1.2.87/tracebloc/raw_mode/survival.py +5 -10
- tracebloc-1.2.85/tracebloc/raw_mode_torch.py → tracebloc-1.2.87/tracebloc/raw_mode/torch_checks.py +11 -15
- tracebloc-1.2.87/tracebloc/raw_mode_frame.py +12 -0
- tracebloc-1.2.87/tracebloc/raw_mode_seed.py +9 -0
- tracebloc-1.2.87/tracebloc/raw_mode_survival.py +18 -0
- tracebloc-1.2.87/tracebloc/raw_mode_torch.py +27 -0
- tracebloc-1.2.87/tracebloc/upload/torch_handler.py +74 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_image_classifier.py +2 -52
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_semantic_segmentation.py +2 -52
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_tabular_classifier.py +3 -53
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_tabular_regression.py +3 -53
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_text_classifier.py +2 -2
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_time_to_event.py +3 -53
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_token_classifier.py +2 -2
- tracebloc-1.2.87/tracebloc/utils/declared_classes.py +6 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/detection_utils.py +2 -2
- tracebloc-1.2.87/tracebloc/validation/estimator_checks.py +16 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/validation/lifelines_checks.py +2 -10
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/validation/scikit_survival_checks.py +2 -10
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/validation/sklearn_checks.py +4 -10
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc.egg-info/PKG-INFO +1 -1
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc.egg-info/SOURCES.txt +15 -1
- tracebloc-1.2.85/tracebloc/raw_mode.py +0 -499
- {tracebloc-1.2.85 → tracebloc-1.2.87}/MANIFEST.in +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/README.md +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/license.txt +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/setup.cfg +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/_deprecation.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/_framework_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/_http_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/_link_answer.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/_link_regeneration.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/_linking.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/_raw_width_note.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/_stored_link.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/_upload_conversion.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/checks.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/client.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/config.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/contracts/stage_urls.v1.json +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/contracts/stage_urls.v1.source.json +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/contracts/survival_payload.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/contracts/training_requirements.source.json +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/contracts/training_requirements.txt +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/acceptance.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/families/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/families/hf/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/families/hf/_build.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/families/hf/_model_type.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/families/tabular.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/families/vision/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/families/vision/_matching.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/gate.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/heads.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/interfaces.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/licence/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/licence/spdx.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/outputs.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/arguments.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/assemble.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/checkpoint.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/gate_rules.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/generate.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/head.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/import_check.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/layer0.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/live.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/model_class.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/remote.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/repo.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/shake.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/owncode/source_tree.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/package.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/report.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/tokenizers.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/convert/training_versions.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/env_login.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/exceptions.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/fetch.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/inference.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/library_fetch.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/lifelines_weights.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/link_model_dataset.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/notebook/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/notebook/templates/_partials.tpl +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/notebook/templates/embeddings.tpl +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/notebook/templates/nlp_finetune.tpl +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/notebook/templates/nlp_generative.tpl +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/notebook/templates/survival.tpl +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/notebook/templates/tabular_timeseries.tpl +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/notebook/templates/vision_from_scratch.tpl +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/seed_weights.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/augmentation.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/callbacks.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/converted.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/features.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/image_channels.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/image_size.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/lora.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/lora_validation.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/loss.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/memory_estimate.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/plan.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/preprocessing.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/shape.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/subdataset.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/submitter.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/training/summary.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/lifelines_time_to_event.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/model_upload.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/scikit_survival_time_to_event.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/sequence_length.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/skl_tabular_base.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/skl_tabular_classifier.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/skl_tabular_regression.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/statistical_base.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/survival_base.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_causal_language_modeler.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_embeddings.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_key_point_detector.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_masked_language_modeler.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_multi_label_text_classifier.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_nlp_base.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_object_detector.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_sentence_pair_classifier.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_seq2seq_modeler.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_time_series_classifier.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/torch_time_series_forecaster.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/weight_loaders.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/upload/weights_parts.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/user.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/class_order.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/constants.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/custom_loss_probe.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/data_probes.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/device.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/entry_point.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/errors.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/feature_dict.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/general_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/key_point_detection_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/model_loading.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/model_params.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/registry.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/resources.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/safetensors_io.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/semantic_segmentation_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/tabular_classification_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/tabular_custom_loss_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/tabular_regression_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/text_classification_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/time_series_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/time_to_event_utils.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/utils/warnings_once.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/validation/__init__.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/validation/checker.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/validation/header_rules.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/validation/pytorch_checks.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc/validation/rewriter.py +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc.egg-info/dependency_links.txt +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc.egg-info/requires.txt +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc.egg-info/top_level.txt +0 -0
- {tracebloc-1.2.85 → tracebloc-1.2.87}/tracebloc_package/__init__.py +0 -0
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "tracebloc"
|
|
7
|
-
version = "1.2.
|
|
7
|
+
version = "1.2.87"
|
|
8
8
|
description = "Build AI with anyone. On data that can't move. SDK for the tracebloc collaborative AI workspace."
|
|
9
9
|
license = "MIT"
|
|
10
10
|
license-files = ["license.txt"]
|
|
@@ -0,0 +1,91 @@
|
|
|
1
|
+
"""Raw mode, the SDK half: check a model that brings its own data preparation.
|
|
2
|
+
|
|
3
|
+
RFC-0243 D9 (backend#5679). A model trained outside tracebloc ships its own
|
|
4
|
+
preparation, e.g. a fitted scikit-learn ``Pipeline`` (imputer + encoder +
|
|
5
|
+
scaler + estimator), and declares it in the model file::
|
|
6
|
+
|
|
7
|
+
DATA_PREPARATION = "model"
|
|
8
|
+
|
|
9
|
+
The backend reads that declaration at an inference-only Start and tells the
|
|
10
|
+
engine, which then gives the model the test table's RAW columns
|
|
11
|
+
(tracebloc/tracebloc-engine#1639):
|
|
12
|
+
|
|
13
|
+
* input: a ``pandas.DataFrame`` of every dataset column except the label, by
|
|
14
|
+
name, in the dataset schema's order, as stored (raw dtypes, categoricals as
|
|
15
|
+
strings, missing values as NaN);
|
|
16
|
+
* output, classification: class NAMES (the dataset's labels) or platform class
|
|
17
|
+
INDICES; output, regression: the target in its original units.
|
|
18
|
+
|
|
19
|
+
The fitted estimator lives in the pickled weights, which the backend never
|
|
20
|
+
unpickles, so the checks that need it run here, at the dataset link, before
|
|
21
|
+
anything is started:
|
|
22
|
+
|
|
23
|
+
* columns: the columns the model was fitted on (``feature_names_in_``) must all
|
|
24
|
+
be among the dataset's non-label columns, since the engine passes them by
|
|
25
|
+
name. A model fitted on a numpy array names none, and is refused;
|
|
26
|
+
* classes (classification): the model's ``classes_`` must be the dataset's
|
|
27
|
+
class names, or all platform class indices, as the engine maps them.
|
|
28
|
+
|
|
29
|
+
That is the scikit-learn tabular pair (tabular classification and regression on
|
|
30
|
+
``sklearn``: scikit-learn, xgboost, lightgbm). The other pairs the engine runs
|
|
31
|
+
raw have their own checks, each the strongest that can run before Start:
|
|
32
|
+
PyTorch tabular and time series in :mod:`tracebloc.raw_mode.torch_checks` (the model's
|
|
33
|
+
own ``preprocess`` / forward called on synthetic rows built from the schema),
|
|
34
|
+
time-to-event on lifelines and scikit-survival in
|
|
35
|
+
:mod:`tracebloc.raw_mode.survival` (the fitted covariates). Any other pair
|
|
36
|
+
(PyTorch time-to-event among them) is the platform's to accept or refuse at
|
|
37
|
+
Start.
|
|
38
|
+
|
|
39
|
+
The package (RFC-0348 C.2): ``declaration`` (the model file's
|
|
40
|
+
``DATA_PREPARATION``), ``pairs`` (the tasks run raw), ``fitted_model`` (the
|
|
41
|
+
weights, unpickled), ``schema`` (the dataset's columns), ``classes`` (class
|
|
42
|
+
mapping), ``sklearn_check``, ``torch_checks`` with ``frame_recording`` and
|
|
43
|
+
``seed_prefixes``, ``survival``, and ``link_check`` (the link's raw-mode step).
|
|
44
|
+
This module re-exports what ``tracebloc/raw_mode.py`` held, so existing callers
|
|
45
|
+
import exactly what they did; the old ``tracebloc.raw_mode_frame``,
|
|
46
|
+
``raw_mode_seed``, ``raw_mode_survival`` and ``raw_mode_torch`` modules re-export
|
|
47
|
+
their moved names too. New code imports from the module named for the concept.
|
|
48
|
+
"""
|
|
49
|
+
|
|
50
|
+
from .declaration import ( # noqa: F401 (re-export, RFC-0348 C.2)
|
|
51
|
+
DATA_PREPARATION,
|
|
52
|
+
MODEL_PREPARES,
|
|
53
|
+
CONTRACT_HEADER,
|
|
54
|
+
_declared,
|
|
55
|
+
declares_raw_mode,
|
|
56
|
+
)
|
|
57
|
+
from .pairs import ( # noqa: F401 (re-export, RFC-0348 C.2)
|
|
58
|
+
SKLEARN_TABULAR,
|
|
59
|
+
TORCH_TABULAR,
|
|
60
|
+
TORCH_SERIES,
|
|
61
|
+
SURVIVAL,
|
|
62
|
+
RAW_MODE_CHECKED,
|
|
63
|
+
)
|
|
64
|
+
from .fitted_model import ( # noqa: F401 (re-export, RFC-0348 C.2)
|
|
65
|
+
load_fitted_model,
|
|
66
|
+
load_beside,
|
|
67
|
+
require_fitted_weights,
|
|
68
|
+
load_or_refuse,
|
|
69
|
+
)
|
|
70
|
+
from .schema import ( # noqa: F401 (re-export, RFC-0348 C.2)
|
|
71
|
+
LABEL_COLUMN,
|
|
72
|
+
SCHEMA_PATH,
|
|
73
|
+
dataset_schema,
|
|
74
|
+
dataset_columns,
|
|
75
|
+
)
|
|
76
|
+
from .classes import ( # noqa: F401 (re-export, RFC-0348 C.2)
|
|
77
|
+
_integral,
|
|
78
|
+
foreign_classes,
|
|
79
|
+
_listed,
|
|
80
|
+
check_exact_classes,
|
|
81
|
+
link_dataset_classes,
|
|
82
|
+
)
|
|
83
|
+
from .sklearn_check import ( # noqa: F401 (re-export, RFC-0348 C.2)
|
|
84
|
+
_FIT_ON_A_DATAFRAME,
|
|
85
|
+
model_feature_names,
|
|
86
|
+
check_raw_model,
|
|
87
|
+
)
|
|
88
|
+
from .link_check import ( # noqa: F401 (re-export, RFC-0348 C.2)
|
|
89
|
+
RawModeCheck,
|
|
90
|
+
raw_mode_input_contract,
|
|
91
|
+
)
|
|
@@ -0,0 +1,100 @@
|
|
|
1
|
+
"""A raw-mode model's classes against the dataset's, mapped as the engine's
|
|
2
|
+
``RawModelAdapter`` maps them: by name first, else as platform class indices."""
|
|
3
|
+
|
|
4
|
+
from __future__ import annotations
|
|
5
|
+
|
|
6
|
+
from ..exceptions import RawModeCheckError
|
|
7
|
+
|
|
8
|
+
|
|
9
|
+
def _integral(values):
|
|
10
|
+
"""``values`` as ints when every one is a whole number, else ``None``
|
|
11
|
+
(the engine's ``_integral_values``: ``1.0`` is ``1``, ``"1"`` is not)."""
|
|
12
|
+
integers = []
|
|
13
|
+
for value in values:
|
|
14
|
+
if isinstance(value, bool) or not isinstance(value, (int, float)):
|
|
15
|
+
try: # numpy scalars
|
|
16
|
+
value = value.item()
|
|
17
|
+
except AttributeError:
|
|
18
|
+
return None
|
|
19
|
+
if isinstance(value, bool) or not isinstance(value, (int, float)):
|
|
20
|
+
return None
|
|
21
|
+
if isinstance(value, float) and not value.is_integer():
|
|
22
|
+
return None
|
|
23
|
+
integers.append(int(value))
|
|
24
|
+
return integers
|
|
25
|
+
|
|
26
|
+
|
|
27
|
+
|
|
28
|
+
def foreign_classes(model_classes, dataset_classes):
|
|
29
|
+
"""The model's classes the engine cannot map onto ``dataset_classes``,
|
|
30
|
+
sorted; ``[]`` when it can.
|
|
31
|
+
|
|
32
|
+
As the engine reads them (``RawModelAdapter``): by NAME first, so an integer
|
|
33
|
+
label that is a class name maps by name; otherwise, when every one is a
|
|
34
|
+
whole number, as platform class indices ``0 .. n-1``.
|
|
35
|
+
"""
|
|
36
|
+
known = {str(name) for name in dataset_classes}
|
|
37
|
+
integral = _integral(model_classes)
|
|
38
|
+
names = (
|
|
39
|
+
[str(value) for value in integral]
|
|
40
|
+
if integral is not None
|
|
41
|
+
else [str(value) for value in model_classes]
|
|
42
|
+
)
|
|
43
|
+
if all(name in known for name in names):
|
|
44
|
+
return []
|
|
45
|
+
if integral is not None and all(0 <= v < len(known) for v in integral):
|
|
46
|
+
return []
|
|
47
|
+
return sorted(set(names) - known)
|
|
48
|
+
|
|
49
|
+
|
|
50
|
+
|
|
51
|
+
def _listed(values):
|
|
52
|
+
"""``classes_`` (a list, a numpy array, a tensor) as a flat list."""
|
|
53
|
+
if hasattr(values, "tolist"):
|
|
54
|
+
values = values.tolist()
|
|
55
|
+
flat = []
|
|
56
|
+
for value in values if isinstance(values, (list, tuple)) else [values]:
|
|
57
|
+
flat.extend(_listed(value) if isinstance(value, (list, tuple)) else [value])
|
|
58
|
+
return flat
|
|
59
|
+
|
|
60
|
+
|
|
61
|
+
|
|
62
|
+
def check_exact_classes(model_classes, dataset_classes):
|
|
63
|
+
"""Refuse, by name, a ``classes_`` that is not EXACTLY the dataset's classes,
|
|
64
|
+
each once (the PyTorch raw paths: a score column per platform class).
|
|
65
|
+
|
|
66
|
+
Mapped as the engine's ``RawModelAdapter`` maps it: by name first, else all
|
|
67
|
+
whole numbers as platform indices ``0 .. n-1``.
|
|
68
|
+
"""
|
|
69
|
+
model_classes = _listed(model_classes)
|
|
70
|
+
known = [str(name) for name in dataset_classes]
|
|
71
|
+
foreign = foreign_classes(model_classes, dataset_classes)
|
|
72
|
+
if foreign:
|
|
73
|
+
raise RawModeCheckError(
|
|
74
|
+
f"The model's classes_ name class(es) {foreign} that are not among "
|
|
75
|
+
f"this dataset's classes {known}."
|
|
76
|
+
)
|
|
77
|
+
integral = _integral(model_classes)
|
|
78
|
+
names = [str(v) for v in (integral if integral is not None else model_classes)]
|
|
79
|
+
if all(name in known for name in names):
|
|
80
|
+
indices = [known.index(name) for name in names]
|
|
81
|
+
else: # all platform indices (foreign_classes passed them)
|
|
82
|
+
indices = integral
|
|
83
|
+
if sorted(indices) != list(range(len(known))):
|
|
84
|
+
raise RawModeCheckError(
|
|
85
|
+
f"The model's classes_ {[str(c) for c in model_classes]} must name "
|
|
86
|
+
f"exactly this dataset's classes {known}, each once, in the model's "
|
|
87
|
+
"output column order."
|
|
88
|
+
)
|
|
89
|
+
|
|
90
|
+
|
|
91
|
+
|
|
92
|
+
def link_dataset_classes(content):
|
|
93
|
+
"""The use case's classes from the link answer: ``class_order``, else the
|
|
94
|
+
keys of ``class_names``."""
|
|
95
|
+
dataset_classes = content.get("class_order")
|
|
96
|
+
if not isinstance(dataset_classes, (list, tuple)) or not dataset_classes:
|
|
97
|
+
class_names = content.get("class_names")
|
|
98
|
+
dataset_classes = list(class_names) if isinstance(class_names, dict) else []
|
|
99
|
+
return list(dataset_classes)
|
|
100
|
+
|
|
@@ -0,0 +1,89 @@
|
|
|
1
|
+
"""Raw mode's declaration: ``DATA_PREPARATION = "model"`` in the model file,
|
|
2
|
+
read as the platform reads it (parsed, never imported)."""
|
|
3
|
+
|
|
4
|
+
from __future__ import annotations
|
|
5
|
+
|
|
6
|
+
import ast
|
|
7
|
+
import os
|
|
8
|
+
import zipfile
|
|
9
|
+
|
|
10
|
+
from ..exceptions import RawModeCheckError
|
|
11
|
+
from ..utils.general_utils import format_exception
|
|
12
|
+
|
|
13
|
+
|
|
14
|
+
#: The module-level constant a model file declares its data preparation with.
|
|
15
|
+
DATA_PREPARATION = "DATA_PREPARATION"
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
#: The declared value meaning "the model brings its own preparation".
|
|
19
|
+
MODEL_PREPARES = "model"
|
|
20
|
+
|
|
21
|
+
|
|
22
|
+
#: The first line of every printed input contract.
|
|
23
|
+
CONTRACT_HEADER = (
|
|
24
|
+
'Raw mode (DATA_PREPARATION = "model"): the model prepares its own data.'
|
|
25
|
+
)
|
|
26
|
+
|
|
27
|
+
|
|
28
|
+
|
|
29
|
+
def _declared(source):
|
|
30
|
+
"""The string ``source`` assigns to ``DATA_PREPARATION`` at module level
|
|
31
|
+
(the last assignment wins, as the platform reads it), or ``None``.
|
|
32
|
+
Parsed, never imported."""
|
|
33
|
+
try:
|
|
34
|
+
tree = ast.parse(source)
|
|
35
|
+
except (SyntaxError, ValueError, RecursionError, MemoryError):
|
|
36
|
+
return None
|
|
37
|
+
value = None
|
|
38
|
+
for node in tree.body:
|
|
39
|
+
if (
|
|
40
|
+
isinstance(node, ast.Assign)
|
|
41
|
+
and len(node.targets) == 1
|
|
42
|
+
and isinstance(node.targets[0], ast.Name)
|
|
43
|
+
and node.targets[0].id == DATA_PREPARATION
|
|
44
|
+
and isinstance(node.value, ast.Constant)
|
|
45
|
+
):
|
|
46
|
+
value = node.value.value
|
|
47
|
+
if not isinstance(value, str):
|
|
48
|
+
return None
|
|
49
|
+
return MODEL_PREPARES if value.strip().lower() == MODEL_PREPARES else "tracebloc"
|
|
50
|
+
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
def declares_raw_mode(model_file_path):
|
|
54
|
+
"""Whether the model file (a ``.py``, or a ``.zip`` of them) declares
|
|
55
|
+
``DATA_PREPARATION = "model"``, as the platform reads it.
|
|
56
|
+
|
|
57
|
+
Raises :class:`RawModeCheckError` when the members of a zip disagree: the
|
|
58
|
+
platform refuses such a model, since it cannot tell which one runs.
|
|
59
|
+
"""
|
|
60
|
+
if not model_file_path or not os.path.isfile(model_file_path):
|
|
61
|
+
return False
|
|
62
|
+
found = {}
|
|
63
|
+
try:
|
|
64
|
+
if zipfile.is_zipfile(model_file_path):
|
|
65
|
+
with zipfile.ZipFile(model_file_path) as archive:
|
|
66
|
+
for info in archive.infolist():
|
|
67
|
+
if info.is_dir() or not info.filename.endswith(".py"):
|
|
68
|
+
continue
|
|
69
|
+
value = _declared(archive.read(info).decode("utf-8-sig", "replace"))
|
|
70
|
+
if value is not None:
|
|
71
|
+
found[info.filename] = value
|
|
72
|
+
else:
|
|
73
|
+
with open(model_file_path, encoding="utf-8-sig", errors="replace") as file:
|
|
74
|
+
value = _declared(file.read())
|
|
75
|
+
if value is not None:
|
|
76
|
+
found[os.path.basename(model_file_path)] = value
|
|
77
|
+
except (OSError, zipfile.BadZipFile) as error:
|
|
78
|
+
raise RawModeCheckError(
|
|
79
|
+
"The model file could not be read to find its data preparation "
|
|
80
|
+
f"declaration: {format_exception(error)}"
|
|
81
|
+
) from error
|
|
82
|
+
if len(set(found.values())) > 1:
|
|
83
|
+
listed = ", ".join(f"{name}: {found[name]}" for name in sorted(found))
|
|
84
|
+
raise RawModeCheckError(
|
|
85
|
+
f"The model's files disagree on its data preparation ({listed}): "
|
|
86
|
+
f'declare DATA_PREPARATION = "model" in one file of the zip, or in none.'
|
|
87
|
+
)
|
|
88
|
+
return MODEL_PREPARES in found.values()
|
|
89
|
+
|
|
@@ -0,0 +1,86 @@
|
|
|
1
|
+
"""The fitted raw-mode model: its uploaded weights, unpickled beside the model
|
|
2
|
+
file, or a named refusal."""
|
|
3
|
+
|
|
4
|
+
from __future__ import annotations
|
|
5
|
+
|
|
6
|
+
import os
|
|
7
|
+
import pickle
|
|
8
|
+
import sys
|
|
9
|
+
import zipfile
|
|
10
|
+
|
|
11
|
+
from ..exceptions import RawModeCheckError
|
|
12
|
+
from ..utils.general_utils import format_exception
|
|
13
|
+
|
|
14
|
+
|
|
15
|
+
def load_fitted_model(weights_path):
|
|
16
|
+
"""The fitted estimator in the weights file (``.joblib``, else pickle)."""
|
|
17
|
+
if os.path.splitext(weights_path)[1].lower() == ".joblib":
|
|
18
|
+
import joblib
|
|
19
|
+
|
|
20
|
+
return joblib.load(weights_path)
|
|
21
|
+
# The user's own weights file, which upload_model already unpickles in this
|
|
22
|
+
# process (``Model.validate_model_file``): no new trust boundary.
|
|
23
|
+
with open(weights_path, "rb") as file:
|
|
24
|
+
return pickle.load(file) # nosec B301
|
|
25
|
+
|
|
26
|
+
|
|
27
|
+
|
|
28
|
+
def load_beside(model_file_path):
|
|
29
|
+
"""``load_fitted_model`` with the model file's folder importable, so a class
|
|
30
|
+
pickled from that file unpickles; a zip model is importable as the archive
|
|
31
|
+
itself. The entry goes to the FRONT of ``sys.path`` for the load only, so a
|
|
32
|
+
folder already on the path (a notebook's own) cannot win with a same-named
|
|
33
|
+
module, and comes out after. Modules imported from those entries are dropped
|
|
34
|
+
after the load, so a later model reads its own. Every raw pair that unpickles
|
|
35
|
+
a fitted model loads through this."""
|
|
36
|
+
if not model_file_path:
|
|
37
|
+
entries = []
|
|
38
|
+
elif zipfile.is_zipfile(model_file_path):
|
|
39
|
+
entries = [os.path.abspath(model_file_path)]
|
|
40
|
+
else:
|
|
41
|
+
entries = [os.path.dirname(os.path.abspath(model_file_path))]
|
|
42
|
+
|
|
43
|
+
def load(weights_path):
|
|
44
|
+
before = set(sys.modules)
|
|
45
|
+
sys.path[:0] = entries
|
|
46
|
+
try:
|
|
47
|
+
return load_fitted_model(weights_path)
|
|
48
|
+
finally:
|
|
49
|
+
for entry in entries:
|
|
50
|
+
sys.path.remove(entry) # the first copy: the one put in front
|
|
51
|
+
for name in set(sys.modules) - before:
|
|
52
|
+
origin = getattr(sys.modules.get(name), "__file__", None)
|
|
53
|
+
if origin and any(
|
|
54
|
+
os.path.abspath(origin).startswith(entry + os.sep)
|
|
55
|
+
for entry in entries
|
|
56
|
+
):
|
|
57
|
+
sys.modules.pop(name, None)
|
|
58
|
+
|
|
59
|
+
return load
|
|
60
|
+
|
|
61
|
+
|
|
62
|
+
|
|
63
|
+
def require_fitted_weights(user):
|
|
64
|
+
"""The uploaded weights file of a model scored as fitted, or a named
|
|
65
|
+
refusal when there is none."""
|
|
66
|
+
weights_path = user._User__weights_file_path
|
|
67
|
+
if not user._User__weights or not weights_path or not os.path.isfile(weights_path):
|
|
68
|
+
raise RawModeCheckError(
|
|
69
|
+
"A model that brings its own data preparation is scored as fitted: "
|
|
70
|
+
"upload it with its fitted weights (weights=True, e.g. "
|
|
71
|
+
"<model>_weights.pkl)."
|
|
72
|
+
)
|
|
73
|
+
return weights_path
|
|
74
|
+
|
|
75
|
+
|
|
76
|
+
|
|
77
|
+
def load_or_refuse(load, weights_path):
|
|
78
|
+
"""``load(weights_path)``; a failure is a named refusal, never a pass."""
|
|
79
|
+
try:
|
|
80
|
+
return load(weights_path)
|
|
81
|
+
except Exception as error:
|
|
82
|
+
raise RawModeCheckError(
|
|
83
|
+
f"Could not load the fitted model from its weights: "
|
|
84
|
+
f"{format_exception(error)}"
|
|
85
|
+
) from error
|
|
86
|
+
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
"""The frame a raw-mode PyTorch model's code is called on at link time
|
|
2
|
-
(:mod:`tracebloc.
|
|
2
|
+
(:mod:`tracebloc.raw_mode.torch_checks`): a DataFrame that records every column the
|
|
3
3
|
code read by name and the dataset does not have, so that read is refused by
|
|
4
4
|
name rather than reported as a failure on the dummy values."""
|
|
5
5
|
|
|
@@ -106,7 +106,8 @@ def reads_frames(model_file_path, category, framework):
|
|
|
106
106
|
columns (the platform's count includes ``time``) and average with it. Raw
|
|
107
107
|
mode is inference only (RFC-0243 D9), so the upload skips the step; the
|
|
108
108
|
link checks the model instead (``check_series_*``, the fitted covariates)."""
|
|
109
|
-
from .
|
|
109
|
+
from .declaration import declares_raw_mode
|
|
110
|
+
from .pairs import SURVIVAL, TORCH_SERIES
|
|
110
111
|
|
|
111
112
|
pairs = TORCH_SERIES | SURVIVAL
|
|
112
113
|
return (category, framework) in pairs and declares_raw_mode(model_file_path)
|
|
@@ -119,7 +120,7 @@ def raw_upload_failure(task):
|
|
|
119
120
|
``("model", error)``, named as the link names it, never raised. The upload
|
|
120
121
|
(``model_func_checks``) and the Run tab's Trains check
|
|
121
122
|
(``tracebloc.checks.trains``) both call this, so they cannot disagree."""
|
|
122
|
-
from
|
|
123
|
+
from ..exceptions import RawModeCheckError
|
|
123
124
|
|
|
124
125
|
path = getattr(task, "_Model__model_path", None)
|
|
125
126
|
try:
|
|
@@ -0,0 +1,58 @@
|
|
|
1
|
+
"""Raw mode's step in the dataset link: pick the pair's check and run it."""
|
|
2
|
+
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
5
|
+
from typing import NamedTuple
|
|
6
|
+
|
|
7
|
+
from .classes import link_dataset_classes
|
|
8
|
+
from .declaration import declares_raw_mode
|
|
9
|
+
from .fitted_model import load_beside, load_or_refuse, require_fitted_weights
|
|
10
|
+
from .pairs import RAW_MODE_CHECKED, SURVIVAL, TORCH_SERIES, TORCH_TABULAR
|
|
11
|
+
from .schema import dataset_columns
|
|
12
|
+
from .sklearn_check import check_raw_model
|
|
13
|
+
|
|
14
|
+
|
|
15
|
+
class RawModeCheck(NamedTuple):
|
|
16
|
+
"""``raw_mode_input_contract``'s answer: the checked input contract (to
|
|
17
|
+
print with the class order) and the fitted model it was checked on."""
|
|
18
|
+
|
|
19
|
+
#: The checked input contract as text.
|
|
20
|
+
contract: str
|
|
21
|
+
#: The fitted model unpickled from the weights (the model that runs), or
|
|
22
|
+
#: ``None`` when the pair's check loads none (the PyTorch pairs).
|
|
23
|
+
fitted: object = None
|
|
24
|
+
|
|
25
|
+
|
|
26
|
+
|
|
27
|
+
def raw_mode_input_contract(user, dataset_id, content):
|
|
28
|
+
"""``User.__check_model``'s raw-mode step, on a passed link answer.
|
|
29
|
+
|
|
30
|
+
``None`` when the model does not declare raw mode, or its task is not one
|
|
31
|
+
the engine runs raw; otherwise a :class:`RawModeCheck`: the checked input
|
|
32
|
+
contract and the fitted model it loaded, which is what runs, so the link's
|
|
33
|
+
federated-averageability gate judges it, not the ``MyModel()`` factory
|
|
34
|
+
(backend#5679, #872). Raises :class:`RawModeCheckError` with a named reason.
|
|
35
|
+
"""
|
|
36
|
+
if not declares_raw_mode(user._User__model_file_path):
|
|
37
|
+
return None
|
|
38
|
+
pair = (user._User__category, user._User__framework)
|
|
39
|
+
if pair not in RAW_MODE_CHECKED:
|
|
40
|
+
return None
|
|
41
|
+
if pair in TORCH_TABULAR or pair in TORCH_SERIES:
|
|
42
|
+
from .torch_checks import torch_input_contract
|
|
43
|
+
|
|
44
|
+
return RawModeCheck(torch_input_contract(user, dataset_id, content))
|
|
45
|
+
if pair in SURVIVAL:
|
|
46
|
+
from .survival import survival_input_contract
|
|
47
|
+
|
|
48
|
+
return RawModeCheck(*survival_input_contract(user, dataset_id))
|
|
49
|
+
weights_path = require_fitted_weights(user)
|
|
50
|
+
model = load_or_refuse(load_beside(user._User__model_file_path), weights_path)
|
|
51
|
+
contract = check_raw_model(
|
|
52
|
+
model,
|
|
53
|
+
columns=dataset_columns(user._User__client, dataset_id),
|
|
54
|
+
category=pair[0],
|
|
55
|
+
dataset_classes=link_dataset_classes(content),
|
|
56
|
+
)
|
|
57
|
+
return RawModeCheck(contract, model)
|
|
58
|
+
|
|
@@ -0,0 +1,54 @@
|
|
|
1
|
+
"""The (category, framework) pairs the engine runs in raw mode."""
|
|
2
|
+
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
5
|
+
from ..utils.constants import (
|
|
6
|
+
LIFELINES_FRAMEWORK,
|
|
7
|
+
PYTORCH_FRAMEWORK,
|
|
8
|
+
SCIKIT_SURVIVAL_FRAMEWORK,
|
|
9
|
+
SKLEARN_FRAMEWORK,
|
|
10
|
+
TABULAR_CLASSIFICATION,
|
|
11
|
+
TABULAR_REGRESSION,
|
|
12
|
+
TIME_SERIES_CLASSIFICATION,
|
|
13
|
+
TIME_SERIES_FORECASTING,
|
|
14
|
+
TIME_TO_EVENT_PREDICTION,
|
|
15
|
+
)
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
#: The (category, framework) pairs the engine runs in raw mode
|
|
19
|
+
#: (tracebloc-engine#1639 sklearn tabular, #1758 PyTorch tabular, #1756 time
|
|
20
|
+
#: series, #1755 time-to-event; backend ``experiments/services/raw_mode.py``).
|
|
21
|
+
SKLEARN_TABULAR = frozenset(
|
|
22
|
+
{
|
|
23
|
+
(TABULAR_CLASSIFICATION, SKLEARN_FRAMEWORK),
|
|
24
|
+
(TABULAR_REGRESSION, SKLEARN_FRAMEWORK),
|
|
25
|
+
}
|
|
26
|
+
)
|
|
27
|
+
|
|
28
|
+
|
|
29
|
+
TORCH_TABULAR = frozenset(
|
|
30
|
+
{
|
|
31
|
+
(TABULAR_CLASSIFICATION, PYTORCH_FRAMEWORK),
|
|
32
|
+
(TABULAR_REGRESSION, PYTORCH_FRAMEWORK),
|
|
33
|
+
}
|
|
34
|
+
)
|
|
35
|
+
|
|
36
|
+
|
|
37
|
+
TORCH_SERIES = frozenset(
|
|
38
|
+
{
|
|
39
|
+
(TIME_SERIES_CLASSIFICATION, PYTORCH_FRAMEWORK),
|
|
40
|
+
(TIME_SERIES_FORECASTING, PYTORCH_FRAMEWORK),
|
|
41
|
+
}
|
|
42
|
+
)
|
|
43
|
+
|
|
44
|
+
|
|
45
|
+
SURVIVAL = frozenset(
|
|
46
|
+
{
|
|
47
|
+
(TIME_TO_EVENT_PREDICTION, LIFELINES_FRAMEWORK),
|
|
48
|
+
(TIME_TO_EVENT_PREDICTION, SCIKIT_SURVIVAL_FRAMEWORK),
|
|
49
|
+
}
|
|
50
|
+
)
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
RAW_MODE_CHECKED = SKLEARN_TABULAR | TORCH_TABULAR | TORCH_SERIES | SURVIVAL
|
|
54
|
+
|
|
@@ -0,0 +1,58 @@
|
|
|
1
|
+
"""The dataset's column schema, as a raw-mode model receives the columns."""
|
|
2
|
+
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
5
|
+
from ..exceptions import APIError, RawModeCheckError
|
|
6
|
+
from ..training.features import _EXCLUDED_COLUMNS
|
|
7
|
+
from ..utils.general_utils import format_exception
|
|
8
|
+
|
|
9
|
+
|
|
10
|
+
#: The engine's label column (``core/data/constants.py``); it never reaches
|
|
11
|
+
#: the model.
|
|
12
|
+
LABEL_COLUMN = "label"
|
|
13
|
+
|
|
14
|
+
|
|
15
|
+
#: Where the dataset's column schema is read (``FeatureConfig.load`` too).
|
|
16
|
+
SCHEMA_PATH = "dataset/global_meta_data/"
|
|
17
|
+
|
|
18
|
+
|
|
19
|
+
|
|
20
|
+
def dataset_schema(client, dataset_id, excluded=_EXCLUDED_COLUMNS):
|
|
21
|
+
"""The dataset's columns and their stored types, in schema order, less the
|
|
22
|
+
ingestor bookkeeping columns ``excluded`` (the label is kept).
|
|
23
|
+
|
|
24
|
+
Raises :class:`RawModeCheckError` when the schema cannot be read or names no
|
|
25
|
+
column besides the label: the check cannot tell, which is not a pass.
|
|
26
|
+
"""
|
|
27
|
+
try:
|
|
28
|
+
response = client.get(
|
|
29
|
+
SCHEMA_PATH, params={"dataset_id": dataset_id}, raise_on_error=False
|
|
30
|
+
)
|
|
31
|
+
schema = response.json().get("schema") if response.status_code == 200 else None
|
|
32
|
+
except Exception as error:
|
|
33
|
+
# raise_on_error=False: only a 426 (SDK below the API floor) has a status.
|
|
34
|
+
if isinstance(error, APIError) and error.status_code != 0:
|
|
35
|
+
raise
|
|
36
|
+
schema, status = None, format_exception(error)
|
|
37
|
+
else:
|
|
38
|
+
status = f"HTTP {response.status_code}"
|
|
39
|
+
columns = {
|
|
40
|
+
column: sql_type
|
|
41
|
+
for column, sql_type in (schema if isinstance(schema, dict) else {}).items()
|
|
42
|
+
if column not in excluded
|
|
43
|
+
}
|
|
44
|
+
if not [column for column in columns if column != LABEL_COLUMN]:
|
|
45
|
+
raise RawModeCheckError(
|
|
46
|
+
f"Could not read dataset '{dataset_id}''s columns ({status}), so a model "
|
|
47
|
+
"that brings its own data preparation cannot be checked against them. "
|
|
48
|
+
"Try again."
|
|
49
|
+
)
|
|
50
|
+
return columns
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
|
|
54
|
+
def dataset_columns(client, dataset_id):
|
|
55
|
+
"""The dataset's non-label columns, in schema order (see
|
|
56
|
+
:func:`dataset_schema`)."""
|
|
57
|
+
return [c for c in dataset_schema(client, dataset_id) if c != LABEL_COLUMN]
|
|
58
|
+
|
tracebloc-1.2.85/tracebloc/raw_mode_seed.py → tracebloc-1.2.87/tracebloc/raw_mode/seed_prefixes.py
RENAMED
|
@@ -32,7 +32,7 @@ def seed_excluded_prefixes(model_file_path):
|
|
|
32
32
|
nothing declares them or the file cannot be read."""
|
|
33
33
|
import tempfile
|
|
34
34
|
|
|
35
|
-
from
|
|
35
|
+
from ..upload.weight_loaders import read_seed_excluded_prefixes
|
|
36
36
|
|
|
37
37
|
if not model_file_path or not os.path.isfile(model_file_path):
|
|
38
38
|
return ()
|