PyIAML 1.0.1__tar.gz → 1.1.0__tar.gz

{pyiaml-1.0.1 → pyiaml-1.1.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: PyIAML
-Version: 1.0.1
+Version: 1.1.0
 Summary: Integrated AutoML for Medical Labs: explainable pipelines for clinical research
 Author-email: Rudy MERIEUX <rmerieux@chu-reims.fr>, Hugo RUELLET <hruellet@chu-reims.fr>, Robin BOURACHOT <rbourachot@chu-reims.fr>, Youssouf DAHLOUK <ydahlouk@chu-reims.fr>
 License:                     GNU GENERAL PUBLIC LICENSE
@@ -713,31 +713,20 @@ Dynamic: license-file
 # IAML — Integrated AutoML for Medical Labs
 
 IAML (Integrated AutoML for Medical Labs) is a Python framework developed by IIAS
-to make machine learning more accessible to clinical research teams. It brings
-preprocessing, model search and evaluation into one workflow for classification,
-regression and survival analysis on tabular data.
+for clinical research teams working with tabular data. It supports classification,
+regression and survival analysis through two complementary strengths:
 
-Researchers can inspect the steps of a selected pipeline, evaluate its predictions
-and generate explanations to discuss with clinicians and data scientists.
-IAML includes a broad set of built-in methods. **Go further with customization.**
-Add your team's preprocessing steps, models, metrics, validation splitters
-and search optimizers to adapt the workflow to your research domain.
-These contributions can be shared and reused across studies.
-The [extension guide](https://iias-research.github.io/iaml/adaptability.html) shows how to get started.
+- **An integrated workflow with very little configuration.** Describe your data,
+  train prediction pipelines with AutoML, evaluate the selected model and request
+  explanations and study outputs. Built-in components and an automatic pipeline
+  let you start with a few lines of Python.
+- **A pipeline API for teams going further.** Compose complete pipelines with
+  `>>`, choose or exclude methods, edit reusable fragments and configure parameter
+  domains. The same API configures metrics, descriptive statistics and explanations,
+  and lets your team's components become reusable building blocks.
 
-## Clinical research workflow
-
-- **Build prediction pipelines:** search preprocessing steps, models and their
-  parameters through a Python API.
-- **Evaluate a study outcome:** choose the metric and validation strategy, then
-  assess the selected candidate on held-out data.
-- **Inspect and explain:** describe pipeline steps, compute SHAP explanations and
-  generate task-specific performance plots.
-- **Document an experiment:** collect method references and optionally retain
-  cross-validation records with `keep_training_history=True`.
-
-The [research guide](https://iias-research.github.io/iaml/scientific.html) shows how to use these outputs when
-reporting a study and recording the settings needed to repeat an experiment.
+Both paths use IAML's training, cross-validation and model selection workflow.
+Start with the defaults or build a recipe around your study's methods.
 
 ## Installation
 
@@ -749,7 +738,7 @@ python -m pip install PyIAML
 
 The distribution is named `PyIAML`. The Python import is `iaml`.
 
-## How to run
+## Use the integrated workflow
 
 Save this example as `example.py` and run it with `python example.py`.
 It uses a dataset bundled with scikit-learn, so no dataset download is needed.
@@ -765,6 +754,7 @@ if __name__ == "__main__":
         X, y, stratify=y, random_state=42
     )
     search = IAML(max_duration=30, max_workers=1)
+    print(search.get_descriptive_statistics(X_train, y_train))
     search.fit(X_train, y_train)
     chosen_model = search.chosen_candidate
     print(chosen_model.evaluate(X_test, y_test))
@@ -776,18 +766,72 @@ malignant and `1` for benign. Keep the `__main__` guard because training uses
 multiprocessing. The example uses one worker and a 30-second search budget.
 Final fitting can take additional time.
 
+Request descriptive statistics, evaluation, plots and explanations when you
+need them. The [worked example](https://iias-research.github.io/iaml/worked_example.html)
+shows the model, performance plots and SHAP outputs produced by this workflow.
+
+## Compose and adapt complete pipelines
+
+Pipeline configuration is optional. To see what it enables, explore
+[Customize IAML](https://iias-research.github.io/iaml/discover_pipelines.html),
+then try the recipe below when your study needs specific methods.
+
+Build a recipe from reusable components, then train it through the same `fit`
+and `evaluate` calls. For example, impute missing values, choose from the
+normalization family except `UnitNormScaler`, and compare two predictors:
+
+```python
+from iaml import IAML
+from iaml.flow import Int, choice, normalizers, use
+from iaml.steps import (
+    LogisticRegression, UnitNormScaler, RandomForestClassifier, SimpleImputer,
+)
+
+pipeline = (
+    use(SimpleImputer).named("cleaning")
+    >> normalizers().remove(UnitNormScaler).named("normalize")
+    >> choice(
+        use(LogisticRegression).named("logistic"),
+        use(
+            RandomForestClassifier,
+            n_estimators=Int(100, 300, initial=150),
+        ).named("forest"),
+    ).named("predictor")
+)
+search = IAML(pipeline=pipeline, max_duration=30, max_workers=1)
+```
+
+Recipes remain editable after construction: navigate by alias, call `add`,
+`remove` or `replace`, and configure all occurrences of a component with
+`find_all(Class).configure(...)`. Clone a fragment to reuse it independently.
+You can also start from `IAML().pipeline` and adapt its visible `main` and
+`minimal` branches.
+
+The [pipeline guide](https://iias-research.github.io/iaml/pipelines/index.html)
+covers construction, editing, fixed values and parameter domains. Continue with
+[study configuration](https://iias-research.github.io/iaml/pipelines/study.html)
+for metrics, descriptive statistics and explanations. The
+[example catalogue](docs/examples/pipelines/README.rst) provides runnable Python examples.
+
+Add custom preprocessing steps, models, metrics, validation splitters or search
+optimizers when your domain needs them. These components can be shared across
+studies; the [extension guide](https://iias-research.github.io/iaml/adaptability.html)
+explains how to implement them.
+
 ## Documentation
 
-The [user guides](https://iias-research.github.io/iaml/) cover data preparation, model search, evaluation
-and interpretation:
+Start with [Discover IAML](https://iias-research.github.io/iaml/) for a visual
+study walkthrough or [Quick Start](https://iias-research.github.io/iaml/quick_start.html)
+to run it. The integrated guides follow the study through
+[training](https://iias-research.github.io/iaml/usage.html),
+[evaluation](https://iias-research.github.io/iaml/evaluation.html),
+[explanations](https://iias-research.github.io/iaml/explainability.html) and
+[reporting](https://iias-research.github.io/iaml/scientific.html).
 
-- [Quick Start](https://iias-research.github.io/iaml/quick_start.html): install IAML and run an example.
-- [01 / Build](https://iias-research.github.io/iaml/usage.html): prepare data and configure a search.
-- [02 / Evaluate](https://iias-research.github.io/iaml/evaluation.html): assess predictions on held-out data.
-- [03 / Explain](https://iias-research.github.io/iaml/explainability.html): inspect methods and interpret feature contributions.
-- [Study reporting](https://iias-research.github.io/iaml/scientific.html): save outputs and record experiment settings.
-- [Extending IAML](https://iias-research.github.io/iaml/adaptability.html): add reusable methods for your team's research.
-- [Component availability](https://iias-research.github.io/iaml/component_status.html): explore the main component families and their API documentation.
+The [advanced guide map](https://iias-research.github.io/iaml/pipelines/index.html)
+connects pipeline construction, study configuration and component extensions.
+Consult [component availability](https://iias-research.github.io/iaml/component_status.html)
+for supported methods and optional dependencies.
 
 ## Credits
 

{pyiaml-1.0.1 → pyiaml-1.1.0}/PyIAML.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: PyIAML
-Version: 1.0.1
+Version: 1.1.0
 Summary: Integrated AutoML for Medical Labs: explainable pipelines for clinical research
 Author-email: Rudy MERIEUX <rmerieux@chu-reims.fr>, Hugo RUELLET <hruellet@chu-reims.fr>, Robin BOURACHOT <rbourachot@chu-reims.fr>, Youssouf DAHLOUK <ydahlouk@chu-reims.fr>
 License:                     GNU GENERAL PUBLIC LICENSE
@@ -713,31 +713,20 @@ Dynamic: license-file
 # IAML — Integrated AutoML for Medical Labs
 
 IAML (Integrated AutoML for Medical Labs) is a Python framework developed by IIAS
-to make machine learning more accessible to clinical research teams. It brings
-preprocessing, model search and evaluation into one workflow for classification,
-regression and survival analysis on tabular data.
+for clinical research teams working with tabular data. It supports classification,
+regression and survival analysis through two complementary strengths:
 
-Researchers can inspect the steps of a selected pipeline, evaluate its predictions
-and generate explanations to discuss with clinicians and data scientists.
-IAML includes a broad set of built-in methods. **Go further with customization.**
-Add your team's preprocessing steps, models, metrics, validation splitters
-and search optimizers to adapt the workflow to your research domain.
-These contributions can be shared and reused across studies.
-The [extension guide](https://iias-research.github.io/iaml/adaptability.html) shows how to get started.
+- **An integrated workflow with very little configuration.** Describe your data,
+  train prediction pipelines with AutoML, evaluate the selected model and request
+  explanations and study outputs. Built-in components and an automatic pipeline
+  let you start with a few lines of Python.
+- **A pipeline API for teams going further.** Compose complete pipelines with
+  `>>`, choose or exclude methods, edit reusable fragments and configure parameter
+  domains. The same API configures metrics, descriptive statistics and explanations,
+  and lets your team's components become reusable building blocks.
 
-## Clinical research workflow
-
-- **Build prediction pipelines:** search preprocessing steps, models and their
-  parameters through a Python API.
-- **Evaluate a study outcome:** choose the metric and validation strategy, then
-  assess the selected candidate on held-out data.
-- **Inspect and explain:** describe pipeline steps, compute SHAP explanations and
-  generate task-specific performance plots.
-- **Document an experiment:** collect method references and optionally retain
-  cross-validation records with `keep_training_history=True`.
-
-The [research guide](https://iias-research.github.io/iaml/scientific.html) shows how to use these outputs when
-reporting a study and recording the settings needed to repeat an experiment.
+Both paths use IAML's training, cross-validation and model selection workflow.
+Start with the defaults or build a recipe around your study's methods.
 
 ## Installation
 
@@ -749,7 +738,7 @@ python -m pip install PyIAML
 
 The distribution is named `PyIAML`. The Python import is `iaml`.
 
-## How to run
+## Use the integrated workflow
 
 Save this example as `example.py` and run it with `python example.py`.
 It uses a dataset bundled with scikit-learn, so no dataset download is needed.
@@ -765,6 +754,7 @@ if __name__ == "__main__":
         X, y, stratify=y, random_state=42
     )
     search = IAML(max_duration=30, max_workers=1)
+    print(search.get_descriptive_statistics(X_train, y_train))
     search.fit(X_train, y_train)
     chosen_model = search.chosen_candidate
     print(chosen_model.evaluate(X_test, y_test))
@@ -776,18 +766,72 @@ malignant and `1` for benign. Keep the `__main__` guard because training uses
 multiprocessing. The example uses one worker and a 30-second search budget.
 Final fitting can take additional time.
 
+Request descriptive statistics, evaluation, plots and explanations when you
+need them. The [worked example](https://iias-research.github.io/iaml/worked_example.html)
+shows the model, performance plots and SHAP outputs produced by this workflow.
+
+## Compose and adapt complete pipelines
+
+Pipeline configuration is optional. To see what it enables, explore
+[Customize IAML](https://iias-research.github.io/iaml/discover_pipelines.html),
+then try the recipe below when your study needs specific methods.
+
+Build a recipe from reusable components, then train it through the same `fit`
+and `evaluate` calls. For example, impute missing values, choose from the
+normalization family except `UnitNormScaler`, and compare two predictors:
+
+```python
+from iaml import IAML
+from iaml.flow import Int, choice, normalizers, use
+from iaml.steps import (
+    LogisticRegression, UnitNormScaler, RandomForestClassifier, SimpleImputer,
+)
+
+pipeline = (
+    use(SimpleImputer).named("cleaning")
+    >> normalizers().remove(UnitNormScaler).named("normalize")
+    >> choice(
+        use(LogisticRegression).named("logistic"),
+        use(
+            RandomForestClassifier,
+            n_estimators=Int(100, 300, initial=150),
+        ).named("forest"),
+    ).named("predictor")
+)
+search = IAML(pipeline=pipeline, max_duration=30, max_workers=1)
+```
+
+Recipes remain editable after construction: navigate by alias, call `add`,
+`remove` or `replace`, and configure all occurrences of a component with
+`find_all(Class).configure(...)`. Clone a fragment to reuse it independently.
+You can also start from `IAML().pipeline` and adapt its visible `main` and
+`minimal` branches.
+
+The [pipeline guide](https://iias-research.github.io/iaml/pipelines/index.html)
+covers construction, editing, fixed values and parameter domains. Continue with
+[study configuration](https://iias-research.github.io/iaml/pipelines/study.html)
+for metrics, descriptive statistics and explanations. The
+[example catalogue](docs/examples/pipelines/README.rst) provides runnable Python examples.
+
+Add custom preprocessing steps, models, metrics, validation splitters or search
+optimizers when your domain needs them. These components can be shared across
+studies; the [extension guide](https://iias-research.github.io/iaml/adaptability.html)
+explains how to implement them.
+
 ## Documentation
 
-The [user guides](https://iias-research.github.io/iaml/) cover data preparation, model search, evaluation
-and interpretation:
+Start with [Discover IAML](https://iias-research.github.io/iaml/) for a visual
+study walkthrough or [Quick Start](https://iias-research.github.io/iaml/quick_start.html)
+to run it. The integrated guides follow the study through
+[training](https://iias-research.github.io/iaml/usage.html),
+[evaluation](https://iias-research.github.io/iaml/evaluation.html),
+[explanations](https://iias-research.github.io/iaml/explainability.html) and
+[reporting](https://iias-research.github.io/iaml/scientific.html).
 
-- [Quick Start](https://iias-research.github.io/iaml/quick_start.html): install IAML and run an example.
-- [01 / Build](https://iias-research.github.io/iaml/usage.html): prepare data and configure a search.
-- [02 / Evaluate](https://iias-research.github.io/iaml/evaluation.html): assess predictions on held-out data.
-- [03 / Explain](https://iias-research.github.io/iaml/explainability.html): inspect methods and interpret feature contributions.
-- [Study reporting](https://iias-research.github.io/iaml/scientific.html): save outputs and record experiment settings.
-- [Extending IAML](https://iias-research.github.io/iaml/adaptability.html): add reusable methods for your team's research.
-- [Component availability](https://iias-research.github.io/iaml/component_status.html): explore the main component families and their API documentation.
+The [advanced guide map](https://iias-research.github.io/iaml/pipelines/index.html)
+connects pipeline construction, study configuration and component extensions.
+Consult [component availability](https://iias-research.github.io/iaml/component_status.html)
+for supported methods and optional dependencies.
 
 ## Credits
 

{pyiaml-1.0.1 → pyiaml-1.1.0}/PyIAML.egg-info/SOURCES.txt

@@ -15,6 +15,7 @@ src/iaml/candidate.py
 src/iaml/core_dispatcher.py
 src/iaml/data_type.py
 src/iaml/dataset.py
+src/iaml/explainers.py
 src/iaml/explanation.py
 src/iaml/iaml.py
 src/iaml/iaml_pipeline.py
@@ -29,6 +30,7 @@ src/iaml/metric_plot.py
 src/iaml/plot.py
 src/iaml/predictor.py
 src/iaml/reference.py
+src/iaml/search_policy.py
 src/iaml/shared_cache.py
 src/iaml/sklearn_preprocessor.py
 src/iaml/stack.py
@@ -36,6 +38,8 @@ src/iaml/statistic.py
 src/iaml/step.py
 src/iaml/step_cache.py
 src/iaml/step_wrapper.py
+src/iaml/steps.py
+src/iaml/study_analyses.py
 src/iaml/timed_pool_executor.py
 src/iaml/type_of_target.py
 src/iaml/void_step.py
@@ -169,6 +173,7 @@ src/iaml/actionables/predictors/regressor/act_sgd_regressor.py
 src/iaml/actionables/predictors/regressor/act_svm_svr.py
 src/iaml/actionables/predictors/regressor/act_xgboost_regressor.py
 src/iaml/actionables/predictors/survival/__init__.py
+src/iaml/actionables/predictors/survival/_survival_forest.py
 src/iaml/actionables/predictors/survival/act_aalen_additive_model.py
 src/iaml/actionables/predictors/survival/act_cox.py
 src/iaml/actionables/predictors/survival/act_coxnet_survival_analysis.py
@@ -184,6 +189,11 @@ src/iaml/decorators/__init__.py
 src/iaml/decorators/all.py
 src/iaml/decorators/is_step.py
 src/iaml/decorators/runner.py
+src/iaml/flow/__init__.py
+src/iaml/flow/compiler.py
+src/iaml/flow/inspection.py
+src/iaml/flow/model.py
+src/iaml/flow/parameters.py
 src/iaml/metrics/__init__.py
 src/iaml/metrics/_classification.py
 src/iaml/metrics/_survival_times.py
@@ -214,6 +224,7 @@ src/iaml/optimizers/genetic_optimizer.py
 src/iaml/optimizers/optimizer.py
 src/iaml/optimizers/random_optimizer.py
 src/iaml/plots/__init__.py
+src/iaml/plots/_classification.py
 src/iaml/plots/bar_plot.py
 src/iaml/plots/box_plot.py
 src/iaml/plots/class_prediction_error_plot.py

pyiaml-1.1.0/README.md

@@ -0,0 +1,129 @@
+# IAML — Integrated AutoML for Medical Labs
+
+IAML (Integrated AutoML for Medical Labs) is a Python framework developed by IIAS
+for clinical research teams working with tabular data. It supports classification,
+regression and survival analysis through two complementary strengths:
+
+- **An integrated workflow with very little configuration.** Describe your data,
+  train prediction pipelines with AutoML, evaluate the selected model and request
+  explanations and study outputs. Built-in components and an automatic pipeline
+  let you start with a few lines of Python.
+- **A pipeline API for teams going further.** Compose complete pipelines with
+  `>>`, choose or exclude methods, edit reusable fragments and configure parameter
+  domains. The same API configures metrics, descriptive statistics and explanations,
+  and lets your team's components become reusable building blocks.
+
+Both paths use IAML's training, cross-validation and model selection workflow.
+Start with the defaults or build a recipe around your study's methods.
+
+## Installation
+
+Use Python 3.10 or later:
+
+```bash
+python -m pip install PyIAML
+```
+
+The distribution is named `PyIAML`. The Python import is `iaml`.
+
+## Use the integrated workflow
+
+Save this example as `example.py` and run it with `python example.py`.
+It uses a dataset bundled with scikit-learn, so no dataset download is needed.
+
+```python
+from sklearn.datasets import load_breast_cancer
+from sklearn.model_selection import train_test_split
+from iaml import IAML
+
+if __name__ == "__main__":
+    X, y = load_breast_cancer(return_X_y=True, as_frame=True)
+    X_train, X_test, y_train, y_test = train_test_split(
+        X, y, stratify=y, random_state=42
+    )
+    search = IAML(max_duration=30, max_workers=1)
+    print(search.get_descriptive_statistics(X_train, y_train))
+    search.fit(X_train, y_train)
+    chosen_model = search.chosen_candidate
+    print(chosen_model.evaluate(X_test, y_test))
+```
+
+`chosen_model` is the selected model, including its preprocessing.
+`evaluate` scores it on the held-out test set. The dataset labels are `0` for
+malignant and `1` for benign. Keep the `__main__` guard because training uses
+multiprocessing. The example uses one worker and a 30-second search budget.
+Final fitting can take additional time.
+
+Request descriptive statistics, evaluation, plots and explanations when you
+need them. The [worked example](https://iias-research.github.io/iaml/worked_example.html)
+shows the model, performance plots and SHAP outputs produced by this workflow.
+
+## Compose and adapt complete pipelines
+
+Pipeline configuration is optional. To see what it enables, explore
+[Customize IAML](https://iias-research.github.io/iaml/discover_pipelines.html),
+then try the recipe below when your study needs specific methods.
+
+Build a recipe from reusable components, then train it through the same `fit`
+and `evaluate` calls. For example, impute missing values, choose from the
+normalization family except `UnitNormScaler`, and compare two predictors:
+
+```python
+from iaml import IAML
+from iaml.flow import Int, choice, normalizers, use
+from iaml.steps import (
+    LogisticRegression, UnitNormScaler, RandomForestClassifier, SimpleImputer,
+)
+
+pipeline = (
+    use(SimpleImputer).named("cleaning")
+    >> normalizers().remove(UnitNormScaler).named("normalize")
+    >> choice(
+        use(LogisticRegression).named("logistic"),
+        use(
+            RandomForestClassifier,
+            n_estimators=Int(100, 300, initial=150),
+        ).named("forest"),
+    ).named("predictor")
+)
+search = IAML(pipeline=pipeline, max_duration=30, max_workers=1)
+```
+
+Recipes remain editable after construction: navigate by alias, call `add`,
+`remove` or `replace`, and configure all occurrences of a component with
+`find_all(Class).configure(...)`. Clone a fragment to reuse it independently.
+You can also start from `IAML().pipeline` and adapt its visible `main` and
+`minimal` branches.
+
+The [pipeline guide](https://iias-research.github.io/iaml/pipelines/index.html)
+covers construction, editing, fixed values and parameter domains. Continue with
+[study configuration](https://iias-research.github.io/iaml/pipelines/study.html)
+for metrics, descriptive statistics and explanations. The
+[example catalogue](docs/examples/pipelines/README.rst) provides runnable Python examples.
+
+Add custom preprocessing steps, models, metrics, validation splitters or search
+optimizers when your domain needs them. These components can be shared across
+studies; the [extension guide](https://iias-research.github.io/iaml/adaptability.html)
+explains how to implement them.
+
+## Documentation
+
+Start with [Discover IAML](https://iias-research.github.io/iaml/) for a visual
+study walkthrough or [Quick Start](https://iias-research.github.io/iaml/quick_start.html)
+to run it. The integrated guides follow the study through
+[training](https://iias-research.github.io/iaml/usage.html),
+[evaluation](https://iias-research.github.io/iaml/evaluation.html),
+[explanations](https://iias-research.github.io/iaml/explainability.html) and
+[reporting](https://iias-research.github.io/iaml/scientific.html).
+
+The [advanced guide map](https://iias-research.github.io/iaml/pipelines/index.html)
+connects pipeline construction, study configuration and component extensions.
+Consult [component availability](https://iias-research.github.io/iaml/component_status.html)
+for supported methods and optional dependencies.
+
+## Credits
+
+- Rudy MERIEUX
+- Robin BOURACHOT
+- Hugo RUELLET
+- Youssouf DAHLOUK

{pyiaml-1.0.1 → pyiaml-1.1.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "PyIAML"
-version = "1.0.1"
+version = "1.1.0"
 authors = [
     { name="Rudy MERIEUX", email="rmerieux@chu-reims.fr" },
     { name="Hugo RUELLET", email="hruellet@chu-reims.fr" },