mudra-ml 0.1.0__tar.gz

mudra_ml-0.1.0/.gitignore

@@ -0,0 +1,38 @@
+# Private working notes, never tracked
+.agent/
+
+# Build artifacts
+dist/
+build/
+*.egg-info/
+.eggs/
+
+# Python
+__pycache__/
+*.py[cod]
+*.so
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+.coverage
+htmlcov/
+coverage.xml
+
+# Environments
+.venv*/
+venv/
+env/
+
+# Credentials
+.pypirc
+
+# Editor and OS
+.vscode/
+.idea/
+.DS_Store
+Thumbs.db
+
+# Generated run output
+mudra_ml_report.*
+run_artifact*
+*.joblib

mudra_ml-0.1.0/CHANGELOG.md

@@ -0,0 +1,29 @@
+# Changelog
+
+All notable changes to this project are recorded here. The format follows
+Keep a Changelog, and the project uses semantic versioning.
+
+## [0.1.0] - 2026-06-04
+
+First public release.
+
+### Added
+- File ingestion for csv, tsv, excel, json, and parquet, with delimiter,
+  encoding, and header auto-detection.
+- Data profiler with per-column type inference (numeric, categorical,
+  datetime, boolean, id, text), missingness, cardinality, distribution stats,
+  and candidate-target ranking.
+- Goal object and rule-based goal inference for task, target, and metric, with
+  operator-set fields taking precedence over inference.
+- Leakage-safe cleaning and preprocessing as a scikit-learn Pipeline and
+  ColumnTransformer: statistical imputation, datetime part extraction, IQR or
+  z-score outlier clipping, one-hot and frequency encoding, and scaling.
+- Rule-based algorithm recommendation keyed on task, dataset size, feature
+  count, cardinality, and operator constraints.
+- Cross-validated training and tuning with RandomizedSearchCV at a fixed seed,
+  task-appropriate evaluation, best-model selection, and feature importance.
+- Markdown and HTML run reports that log every decision and the rule behind it.
+- Model and pipeline persistence through joblib, with a predict path on loaded
+  artifacts.
+- Command line interface with `run` and `profile`.
+- Optional xgboost and lightgbm candidates through the `boost` extra.

mudra_ml-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Mudit Nautiyal
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

mudra_ml-0.1.0/PKG-INFO

@@ -0,0 +1,213 @@
+Metadata-Version: 2.4
+Name: mudra-ml
+Version: 0.1.0
+Summary: Glass-box autonomous data science: profile, clean, model, and explain every decision.
+Project-URL: Homepage, https://github.com/MuditNautiyal-21/MudraML
+Project-URL: Repository, https://github.com/MuditNautiyal-21/MudraML
+Project-URL: Issues, https://github.com/MuditNautiyal-21/MudraML/issues
+Project-URL: Changelog, https://github.com/MuditNautiyal-21/MudraML/blob/main/CHANGELOG.md
+Author: Mudit Nautiyal
+License: MIT License
+        
+        Copyright (c) 2026 Mudit Nautiyal
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+License-File: LICENSE
+Keywords: automl,data-science,explainable,machine-learning,pipeline,scikit-learn
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Typing :: Typed
+Requires-Python: >=3.10
+Requires-Dist: jinja2>=3.1
+Requires-Dist: joblib>=1.2
+Requires-Dist: numpy>=1.23
+Requires-Dist: pandas>=1.5
+Requires-Dist: scikit-learn>=1.2
+Requires-Dist: typer>=0.9
+Provides-Extra: boost
+Requires-Dist: lightgbm>=4.0; extra == 'boost'
+Requires-Dist: xgboost>=1.7; extra == 'boost'
+Provides-Extra: dev
+Requires-Dist: build>=1.0; extra == 'dev'
+Requires-Dist: mypy>=1.8; extra == 'dev'
+Requires-Dist: openpyxl>=3.1; extra == 'dev'
+Requires-Dist: pyarrow>=12.0; extra == 'dev'
+Requires-Dist: pytest-cov>=4.1; extra == 'dev'
+Requires-Dist: pytest>=7.4; extra == 'dev'
+Requires-Dist: ruff>=0.4; extra == 'dev'
+Requires-Dist: twine>=5.0; extra == 'dev'
+Provides-Extra: excel
+Requires-Dist: openpyxl>=3.1; extra == 'excel'
+Provides-Extra: files
+Requires-Dist: openpyxl>=3.1; extra == 'files'
+Requires-Dist: pyarrow>=12.0; extra == 'files'
+Provides-Extra: parquet
+Requires-Dist: pyarrow>=12.0; extra == 'parquet'
+Description-Content-Type: text/markdown
+
+# MudraML
+
+Automated, glass-box data science. Point it at a data file, get a fitted model and a report of every decision behind it.
+
+[![PyPI version](https://img.shields.io/pypi/v/mudra-ml.svg)](https://pypi.org/project/mudra-ml/)
+[![Python versions](https://img.shields.io/pypi/pyversions/mudra-ml.svg)](https://pypi.org/project/mudra-ml/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+[![CI](https://github.com/MuditNautiyal-21/MudraML/actions/workflows/ci.yml/badge.svg)](https://github.com/MuditNautiyal-21/MudraML/actions/workflows/ci.yml)
+
+MudraML automates the common data science workflow and shows its work. You point it at a data file, optionally state a goal, and it ingests the data, profiles it, cleans it, picks an algorithm, trains and tunes a shortlist of models, evaluates them, and returns the best fitted model together with a report of every decision it made and the rule behind that decision.
+
+The point of difference is the decision engine. It is rule-based and statistical, not another model. Outlier handling uses IQR or z-score rules. Missing values are filled by median, mode, or a constant, or the column is dropped past a missingness threshold. The algorithm shortlist comes from a documented rule set keyed on the task, the dataset size, the feature count, and your constraints. Every one of those choices is written into the report, so a person can read why the pipeline did what it did and disagree with it if they want.
+
+This is the glass-box position: the models are the product, and the way the pipeline reaches them is auditable rather than hidden inside a search.
+
+## Install
+
+```
+pip install mudra-ml
+```
+
+Optional extras:
+
+```
+pip install mudra-ml[files]    # parquet and excel readers
+pip install mudra-ml[boost]    # xgboost and lightgbm candidates
+```
+
+The library runs fully on the scikit-learn core. The boosters are added to the shortlist only when the extra is installed.
+
+## Quickstart
+
+Fully automatic. MudraML infers the task, the target, and the metric:
+
+```python
+from mudra_ml import Mudra
+
+m = Mudra()
+result = m.run("data.csv")
+print(result.report_path)   # markdown and HTML report on disk
+model = result.best_model   # fitted, ready to predict
+```
+
+Operator-defined goal. You set what you care about and MudraML honors it:
+
+```python
+result = m.run(
+    "churn.csv",
+    target="churn",
+    task="classification",
+    metric="f1",
+    constraints={"interpretable": True, "max_train_seconds": 120},
+)
+```
+
+When `interpretable` is set, the shortlist is limited to models you can read directly, such as logistic regression and a single decision tree. The report states which goal fields you set and which ones were inferred.
+
+## What the report looks like
+
+Every run writes a report that records each decision and the rule that produced it. An excerpt:
+
+```
+GOAL
+  task    classification   (inferred: target has 2 unique values)
+  target  churn            (operator-set)
+  metric  f1               (default for classification)
+
+PROFILE
+  18 columns: 12 numeric, 5 categorical, 1 datetime
+  dropped customer_id (id column, 100 percent unique)
+  missing: tenure 3.2 percent, region 0.1 percent
+
+PREPROCESS  (fit on train split only)
+  tenure        median imputation         (3.2 percent missing, below 40 percent drop threshold)
+  region        most-frequent imputation, one-hot   (4 categories, low cardinality)
+  signup_date   extracted year, month, day-of-week
+  numeric       standard scaling
+
+RECOMMEND
+  shortlist     logistic regression, random forest, gradient boosting
+  rule          classification, 5000 rows, 23 features after encoding, no interpretable constraint
+
+EVALUATE  (held-out test)
+  best          gradient boosting   f1 0.81
+  also          random forest 0.79, logistic regression 0.74
+  top features  tenure, monthly_charges, contract_type
+```
+
+The numbers above are illustrative. Your run writes the real values for your data.
+
+## Predict and reuse
+
+```python
+result.save("run_artifact")            # pipeline + model + metadata
+loaded = Mudra.load("run_artifact")
+preds = loaded.predict(new_dataframe)
+```
+
+The preprocessing pipeline travels with the model, so new rows are transformed the same way the training rows were.
+
+## Command line
+
+```
+mudra-ml run data.csv --target churn --task classification --metric f1
+mudra-ml profile data.csv
+```
+
+`run` writes the report and prints the selected model and its held-out metrics. `profile` prints the inferred column types, missingness, cardinality, and the candidate target columns.
+
+## What it does, stage by stage
+
+1. Ingest. Readers for csv, tsv, excel, json, and parquet. For delimited text the delimiter, encoding, and header row are detected.
+2. Profile. Per-column type inference (numeric, categorical, datetime, boolean, id, text), missingness, cardinality, distribution stats, and candidate-target ranking.
+3. Goal. Rule-based inference of the task, target, and metric, with any field you set taking precedence.
+4. Preprocess. A leakage-safe scikit-learn Pipeline and ColumnTransformer. Imputation, datetime part extraction, outlier clipping, encoding, and scaling are all fit on the training split only.
+5. Recommend. A documented rule set returns a candidate shortlist.
+6. Train and evaluate. Cross-validated training, tuning with RandomizedSearchCV at a fixed seed, held-out scoring, best-model selection, and feature importance where the model exposes it.
+7. Report. Markdown and HTML that log every decision and the rule that produced it.
+
+## Why leakage safety matters here
+
+Every statistic that preprocessing needs (a median, a category frequency, an outlier bound, a scaler mean) is learned during `fit`. MudraML fits the pipeline on the training split and only transforms the test split. No information from the test data reaches the model through preprocessing. The test suite checks this property directly: it fits on a slice with a known mean and confirms the learned imputation value matches the train slice rather than the whole dataset.
+
+## Determinism
+
+One `random_state` is threaded through every stochastic step (the split, the search, the estimators) and defaults to a fixed value. Two runs on the same data and the same goal produce the same result and the same report.
+
+## Tasks and metrics
+
+| Task | Default metric | Also reported |
+| --- | --- | --- |
+| classification | f1 | accuracy, precision, recall, roc_auc, confusion matrix |
+| regression | rmse | mae, mse, r2 |
+| clustering | silhouette | davies_bouldin |
+
+## Scope
+
+This release covers the supervised classification and regression cases and KMeans clustering, end to end, with the decision log and the report. Deep text modeling, time series, model-based imputation, and a search beyond curated grids are out of scope by design, since the engine is meant to stay explainable. See the changelog for the version history.
+
+## License
+
+MIT. See [LICENSE](LICENSE).

mudra_ml-0.1.0/README.md

@@ -0,0 +1,142 @@
+# MudraML
+
+Automated, glass-box data science. Point it at a data file, get a fitted model and a report of every decision behind it.
+
+[![PyPI version](https://img.shields.io/pypi/v/mudra-ml.svg)](https://pypi.org/project/mudra-ml/)
+[![Python versions](https://img.shields.io/pypi/pyversions/mudra-ml.svg)](https://pypi.org/project/mudra-ml/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+[![CI](https://github.com/MuditNautiyal-21/MudraML/actions/workflows/ci.yml/badge.svg)](https://github.com/MuditNautiyal-21/MudraML/actions/workflows/ci.yml)
+
+MudraML automates the common data science workflow and shows its work. You point it at a data file, optionally state a goal, and it ingests the data, profiles it, cleans it, picks an algorithm, trains and tunes a shortlist of models, evaluates them, and returns the best fitted model together with a report of every decision it made and the rule behind that decision.
+
+The point of difference is the decision engine. It is rule-based and statistical, not another model. Outlier handling uses IQR or z-score rules. Missing values are filled by median, mode, or a constant, or the column is dropped past a missingness threshold. The algorithm shortlist comes from a documented rule set keyed on the task, the dataset size, the feature count, and your constraints. Every one of those choices is written into the report, so a person can read why the pipeline did what it did and disagree with it if they want.
+
+This is the glass-box position: the models are the product, and the way the pipeline reaches them is auditable rather than hidden inside a search.
+
+## Install
+
+```
+pip install mudra-ml
+```
+
+Optional extras:
+
+```
+pip install mudra-ml[files]    # parquet and excel readers
+pip install mudra-ml[boost]    # xgboost and lightgbm candidates
+```
+
+The library runs fully on the scikit-learn core. The boosters are added to the shortlist only when the extra is installed.
+
+## Quickstart
+
+Fully automatic. MudraML infers the task, the target, and the metric:
+
+```python
+from mudra_ml import Mudra
+
+m = Mudra()
+result = m.run("data.csv")
+print(result.report_path)   # markdown and HTML report on disk
+model = result.best_model   # fitted, ready to predict
+```
+
+Operator-defined goal. You set what you care about and MudraML honors it:
+
+```python
+result = m.run(
+    "churn.csv",
+    target="churn",
+    task="classification",
+    metric="f1",
+    constraints={"interpretable": True, "max_train_seconds": 120},
+)
+```
+
+When `interpretable` is set, the shortlist is limited to models you can read directly, such as logistic regression and a single decision tree. The report states which goal fields you set and which ones were inferred.
+
+## What the report looks like
+
+Every run writes a report that records each decision and the rule that produced it. An excerpt:
+
+```
+GOAL
+  task    classification   (inferred: target has 2 unique values)
+  target  churn            (operator-set)
+  metric  f1               (default for classification)
+
+PROFILE
+  18 columns: 12 numeric, 5 categorical, 1 datetime
+  dropped customer_id (id column, 100 percent unique)
+  missing: tenure 3.2 percent, region 0.1 percent
+
+PREPROCESS  (fit on train split only)
+  tenure        median imputation         (3.2 percent missing, below 40 percent drop threshold)
+  region        most-frequent imputation, one-hot   (4 categories, low cardinality)
+  signup_date   extracted year, month, day-of-week
+  numeric       standard scaling
+
+RECOMMEND
+  shortlist     logistic regression, random forest, gradient boosting
+  rule          classification, 5000 rows, 23 features after encoding, no interpretable constraint
+
+EVALUATE  (held-out test)
+  best          gradient boosting   f1 0.81
+  also          random forest 0.79, logistic regression 0.74
+  top features  tenure, monthly_charges, contract_type
+```
+
+The numbers above are illustrative. Your run writes the real values for your data.
+
+## Predict and reuse
+
+```python
+result.save("run_artifact")            # pipeline + model + metadata
+loaded = Mudra.load("run_artifact")
+preds = loaded.predict(new_dataframe)
+```
+
+The preprocessing pipeline travels with the model, so new rows are transformed the same way the training rows were.
+
+## Command line
+
+```
+mudra-ml run data.csv --target churn --task classification --metric f1
+mudra-ml profile data.csv
+```
+
+`run` writes the report and prints the selected model and its held-out metrics. `profile` prints the inferred column types, missingness, cardinality, and the candidate target columns.
+
+## What it does, stage by stage
+
+1. Ingest. Readers for csv, tsv, excel, json, and parquet. For delimited text the delimiter, encoding, and header row are detected.
+2. Profile. Per-column type inference (numeric, categorical, datetime, boolean, id, text), missingness, cardinality, distribution stats, and candidate-target ranking.
+3. Goal. Rule-based inference of the task, target, and metric, with any field you set taking precedence.
+4. Preprocess. A leakage-safe scikit-learn Pipeline and ColumnTransformer. Imputation, datetime part extraction, outlier clipping, encoding, and scaling are all fit on the training split only.
+5. Recommend. A documented rule set returns a candidate shortlist.
+6. Train and evaluate. Cross-validated training, tuning with RandomizedSearchCV at a fixed seed, held-out scoring, best-model selection, and feature importance where the model exposes it.
+7. Report. Markdown and HTML that log every decision and the rule that produced it.
+
+## Why leakage safety matters here
+
+Every statistic that preprocessing needs (a median, a category frequency, an outlier bound, a scaler mean) is learned during `fit`. MudraML fits the pipeline on the training split and only transforms the test split. No information from the test data reaches the model through preprocessing. The test suite checks this property directly: it fits on a slice with a known mean and confirms the learned imputation value matches the train slice rather than the whole dataset.
+
+## Determinism
+
+One `random_state` is threaded through every stochastic step (the split, the search, the estimators) and defaults to a fixed value. Two runs on the same data and the same goal produce the same result and the same report.
+
+## Tasks and metrics
+
+| Task | Default metric | Also reported |
+| --- | --- | --- |
+| classification | f1 | accuracy, precision, recall, roc_auc, confusion matrix |
+| regression | rmse | mae, mse, r2 |
+| clustering | silhouette | davies_bouldin |
+
+## Scope
+
+This release covers the supervised classification and regression cases and KMeans clustering, end to end, with the decision log and the report. Deep text modeling, time series, model-based imputation, and a search beyond curated grids are out of scope by design, since the engine is meant to stay explainable. See the changelog for the version history.
+
+## License
+
+MIT. See [LICENSE](LICENSE).

mudra_ml-0.1.0/pyproject.toml

@@ -0,0 +1,106 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "mudra-ml"
+version = "0.1.0"
+description = "Glass-box autonomous data science: profile, clean, model, and explain every decision."
+readme = "README.md"
+requires-python = ">=3.10"
+license = { file = "LICENSE" }
+authors = [{ name = "Mudit Nautiyal" }]
+keywords = [
+    "automl",
+    "machine-learning",
+    "data-science",
+    "scikit-learn",
+    "explainable",
+    "pipeline",
+]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Science/Research",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Operating System :: OS Independent",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Typing :: Typed",
+]
+dependencies = [
+    "pandas>=1.5",
+    "numpy>=1.23",
+    "scikit-learn>=1.2",
+    "joblib>=1.2",
+    "jinja2>=3.1",
+    "typer>=0.9",
+]
+
+[project.optional-dependencies]
+parquet = ["pyarrow>=12.0"]
+excel = ["openpyxl>=3.1"]
+boost = ["xgboost>=1.7", "lightgbm>=4.0"]
+files = ["pyarrow>=12.0", "openpyxl>=3.1"]
+dev = [
+    "pytest>=7.4",
+    "pytest-cov>=4.1",
+    "ruff>=0.4",
+    "mypy>=1.8",
+    "build>=1.0",
+    "twine>=5.0",
+    "pyarrow>=12.0",
+    "openpyxl>=3.1",
+]
+
+[project.urls]
+Homepage = "https://github.com/MuditNautiyal-21/MudraML"
+Repository = "https://github.com/MuditNautiyal-21/MudraML"
+Issues = "https://github.com/MuditNautiyal-21/MudraML/issues"
+Changelog = "https://github.com/MuditNautiyal-21/MudraML/blob/main/CHANGELOG.md"
+
+[project.scripts]
+mudra-ml = "mudra_ml.cli:app"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/mudra_ml"]
+
+[tool.hatch.build.targets.sdist]
+include = [
+    "src/mudra_ml",
+    "tests",
+    "README.md",
+    "CHANGELOG.md",
+    "LICENSE",
+    "pyproject.toml",
+]
+
+[tool.ruff]
+line-length = 100
+target-version = "py310"
+src = ["src", "tests"]
+
+[tool.ruff.lint]
+select = ["E", "F", "I", "UP", "B", "W"]
+ignore = ["B008"]
+
+[tool.ruff.lint.per-file-ignores]
+# The report module holds long inline HTML and markdown templates.
+"src/mudra_ml/report.py" = ["E501"]
+"tests/*" = ["E501"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+addopts = "-q"
+
+[tool.coverage.run]
+source = ["mudra_ml"]
+branch = true
+
+[tool.mypy]
+python_version = "3.10"
+ignore_missing_imports = true
+warn_unused_ignores = false

mudra_ml-0.1.0/src/mudra_ml/__init__.py

@@ -0,0 +1,34 @@
+"""MudraML: glass-box autonomous data science.
+
+The decision engine that drives the pipeline is rule-based and statistical.
+It is deterministic, logged, and explainable. The machine learning models are
+the output it produces, not the mechanism by which it chooses what to do.
+"""
+
+from __future__ import annotations
+
+from .core import Mudra, RunResult
+from .evaluate import evaluate
+from .goal import Goal, infer_goal
+from .ingest import load
+from .preprocess import build_pipeline
+from .profile import DataProfile, DataProfiler
+from .recommend import recommend_models
+from .report import write_report
+
+__version__ = "0.1.0"
+
+__all__ = [
+    "Mudra",
+    "RunResult",
+    "Goal",
+    "infer_goal",
+    "load",
+    "DataProfiler",
+    "DataProfile",
+    "build_pipeline",
+    "recommend_models",
+    "evaluate",
+    "write_report",
+    "__version__",
+]