Bugfixing & improvements

.github/workflows/test_R.yml

@@ -34,7 +34,7 @@ jobs:
           Rscript -e "install.packages(c('remotes','reticulate'))"
       - name: Test R
         run: |
-          Rscript tests/bindings/R/test_survival_analysis.R
           Rscript tests/bindings/R/test_classification.R
           Rscript tests/bindings/R/test_classification_with_missing_data.R
           Rscript tests/bindings/R/test_regression.R
+          Rscript tests/bindings/R/test_survival_analysis.R

src/autoprognosis/plugins/core/base_plugin.py

@@ -42,6 +42,7 @@ def __init__(self) -> None:
         self.output = pd.DataFrame
         self._backup_encoders: Optional[Dict[str, LabelEncoder]] = {}
         self._drop_features: Optional[List[str]] = []
+        self._fitted = False
 
     def change_output(self, output: str) -> None:
         if output not in ["pandas", "numpy"]:
@@ -124,6 +125,12 @@ def subtype() -> str:
     def fqdn(cls) -> str:
         return cls.type() + "." + cls.subtype() + "." + cls.name()
 
+    def is_fitted(self) -> bool:
+        try:
+            return self._fitted
+        except BaseException:
+            return True
+
     def fit_transform(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> pd.DataFrame:
         return pd.DataFrame(self.fit(X, *args, *kwargs).transform(X))
 
@@ -165,13 +172,19 @@ def _transform_input(self, X: pd.DataFrame) -> pd.DataFrame:
     def fit(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> "Plugin":
         X = self._fit_input(X)
 
-        return self._fit(X, *args, **kwargs)
+        self._fit(X, *args, **kwargs)
+
+        self._fitted = True
+
+        return self
 
     @abstractmethod
     def _fit(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> "Plugin":
         ...
 
     def transform(self, X: pd.DataFrame) -> pd.DataFrame:
+        if not self.is_fitted():
+            raise RuntimeError("Fit the model first")
         X = self._transform_input(X)
         return self.output(self._transform(X))
 
@@ -180,6 +193,8 @@ def _transform(self, X: pd.DataFrame) -> pd.DataFrame:
         ...
 
     def predict(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> pd.DataFrame:
+        if not self.is_fitted():
+            raise RuntimeError("Fit the model first")
         X = self._transform_input(X)
         return self.output(self._predict(X, *args, *kwargs))
 

src/autoprognosis/plugins/ensemble/classifiers.py

@@ -13,7 +13,9 @@
 import autoprognosis.logger as log
 from autoprognosis.plugins.ensemble.combos import SimpleClassifierAggregator, Stacking
 from autoprognosis.plugins.explainers import Explainers
+from autoprognosis.plugins.imputers import Imputers
 from autoprognosis.plugins.pipeline import Pipeline, PipelineMeta
+from autoprognosis.plugins.prediction.classifiers import Classifiers
 from autoprognosis.utils.parallel import cpu_count
 import autoprognosis.utils.serialization as serialization
 from autoprognosis.utils.tester import classifier_evaluator
@@ -91,12 +93,22 @@ def __init__(
         self.explanations_nepoch = explanations_nepoch
         self.explainers = explainers
 
+        self._fitted = True
+        for model in models:
+            self._fitted = self._fitted or model.is_fitted()
+
         for idx, weight in enumerate(weights):
             if weight == 0:
                 continue
             self.models.append(models[idx])
             self.weights.append(weights[idx])
 
+    def is_fitted(self) -> bool:
+        try:
+            return self._fitted
+        except BaseException:
+            return True  # backwards compatible
+
     def fit(self, X: pd.DataFrame, Y: pd.DataFrame) -> "WeightedEnsemble":
         def fit_model(k: int) -> Any:
             return self.models[k].fit(X, Y)
@@ -121,9 +133,13 @@ def fit_model(k: int) -> Any:
             )
             self.explainers[exp] = exp_model
 
+        self._fitted = True
         return self
 
     def predict_proba(self, X: pd.DataFrame, *args: Any) -> pd.DataFrame:
+        if not self.is_fitted():
+            raise RuntimeError("Fit the model first")
+
         preds_ = []
         for k in range(len(self.models)):
             preds_.append(self.models[k].predict_proba(X, *args) * self.weights[k])
@@ -307,22 +323,31 @@ class StackingEnsemble(BaseEnsemble):
     def __init__(
         self,
         models: List[PipelineMeta],
-        meta_model: PipelineMeta = Pipeline(
-            ["prediction.classifier.logistic_regression"]
-        )(output="numpy"),
+        meta_model: Optional[PipelineMeta] = None,
         clf: Union[None, Stacking] = None,
         explainer_plugins: list = [],
         explanations_nepoch: int = 10000,
     ) -> None:
         super().__init__()
 
         self.models = models
+        if meta_model is None:
+            meta_model = Pipeline(
+                [
+                    Imputers().get_type("ice").fqdn(),
+                    Classifiers().get_type("logistic_regression").fqdn(),
+                ]
+            )(output="numpy")
         self.meta_model = meta_model
 
         self.explainer_plugins = explainer_plugins
         self.explainers: Optional[dict]
         self.explanations_nepoch = explanations_nepoch
 
+        self._fitted = True
+        for model in models:
+            self._fitted = self._fitted or model.is_fitted()
+
         for model in self.models:
             model.change_output("numpy")
 
@@ -335,6 +360,12 @@ def __init__(
                 use_proba=True,
             )
 
+    def is_fitted(self) -> bool:
+        try:
+            return self._fitted
+        except BaseException:
+            return True  # backwards compatible
+
     def fit(self, X: pd.DataFrame, Y: pd.DataFrame) -> "StackingEnsemble":
         self.clf.fit(X, Y)
 
@@ -349,10 +380,13 @@ def fit(self, X: pd.DataFrame, Y: pd.DataFrame) -> "StackingEnsemble":
                 n_epoch=self.explanations_nepoch,
                 prefit=True,
             )
-
+        self._fitted = True
         return self
 
     def predict_proba(self, X: pd.DataFrame, *args: Any) -> pd.DataFrame:
+        if not self.is_fitted():
+            raise RuntimeError("Fit the model first")
+
         return pd.DataFrame(self.clf.predict_proba(X))
 
     def explain(self, X: pd.DataFrame, *args: Any) -> pd.DataFrame:
@@ -428,11 +462,21 @@ def __init__(
         self.explainers: Optional[dict]
         self.explanations_nepoch = explanations_nepoch
 
+        self._fitted = True
+        for model in models:
+            self._fitted = self._fitted or model.is_fitted()
+
         if clf:
             self.clf = clf
         else:
             self.clf = SimpleClassifierAggregator(models, method=method)
 
+    def is_fitted(self) -> bool:
+        try:
+            return self._fitted
+        except BaseException:
+            return True  # backwards compatible
+
     def fit(self, X: pd.DataFrame, Y: pd.DataFrame) -> "AggregatingEnsemble":
         Y = pd.DataFrame(Y).values.ravel()
 
@@ -449,10 +493,13 @@ def fit(self, X: pd.DataFrame, Y: pd.DataFrame) -> "AggregatingEnsemble":
                 n_epoch=self.explanations_nepoch,
                 prefit=True,
             )
-
+        self._fitted = True
         return self
 
     def predict_proba(self, X: pd.DataFrame, *args: Any) -> pd.DataFrame:
+        if not self.is_fitted():
+            raise RuntimeError("Fit the model first")
+
         return pd.DataFrame(self.clf.predict_proba(X))
 
     def explain(self, X: pd.DataFrame, *args: Any) -> pd.DataFrame:

src/autoprognosis/plugins/ensemble/combos.py

@@ -16,7 +16,10 @@
 import numpy as np
 from pyod.utils.utility import check_parameter
 from scipy.special import erf
+from sklearn.experimental import enable_iterative_imputer  # noqa: F401,E402
+from sklearn.impute import IterativeImputer
 from sklearn.linear_model import LogisticRegression
+from sklearn.pipeline import Pipeline
 from sklearn.preprocessing import MinMaxScaler
 from sklearn.utils import (
     check_array,
@@ -417,9 +420,6 @@ class Stacking(BaseAggregator):
     base_estimators: list or numpy array (n_estimators,)
         A list of base classifiers.
 
-    meta_clf : object, optional (default=LogisticRegression)
-        The meta classifier to make the final prediction.
-
     n_folds : int, optional (default=2)
         The number of splits of the training sample.
 
@@ -451,7 +451,7 @@ def __init__(
         self,
         base_estimators,
         meta_clf=None,
-        n_folds=2,
+        n_folds=3,
         keep_original=True,
         use_proba=False,
         shuffle_data=False,
@@ -473,7 +473,9 @@ def __init__(
         if meta_clf is not None:
             self.meta_clf = meta_clf
         else:
-            self.meta_clf = LogisticRegression()
+            self.meta_clf = Pipeline(
+                ("imputer", IterativeImputer()), ("output", LogisticRegression())
+            )
 
         # set flags
         self.keep_original = keep_original
@@ -504,8 +506,8 @@ def fit(self, X, y):
         """
 
         # Validate inputs X and y
-        X, y = check_X_y(X, y)
-        X = check_array(X)
+        X, y = check_X_y(X, y, force_all_finite=False)
+        X = check_array(X, force_all_finite=False)
         self._set_n_classes(y)
 
         n_samples = X.shape[0]
@@ -574,7 +576,7 @@ def _process_data(self, X):
             The processed dataset of X.
         """
         check_is_fitted(self, ["fitted_"])
-        X = check_array(X)
+        X = check_array(X, force_all_finite=False)
         n_samples = X.shape[0]
 
         # initialize matrix for storing newly generated features
@@ -718,8 +720,8 @@ def fit(self, X, y):
         """
 
         # Validate inputs X and y
-        X, y = check_X_y(X, y)
-        X = check_array(X)
+        X, y = check_X_y(X, y, force_all_finite=False)
+        X = check_array(X, force_all_finite=False)
         self._set_n_classes(y)
 
         if self.pre_fitted:
@@ -744,7 +746,7 @@ def predict(self, X):
         labels : numpy array of shape (n_samples,)
             Class labels for each data sample.
         """
-        X = check_array(X)
+        X = check_array(X, force_all_finite=False)
 
         all_scores = np.zeros([X.shape[0], self.n_base_estimators_])
 
@@ -782,7 +784,7 @@ def predict_proba(self, X):
             The class probabilities of the input samples.
             Classes are ordered by lexicographic order.
         """
-        X = check_array(X)
+        X = check_array(X, force_all_finite=False)
         all_scores = np.zeros([X.shape[0], self._classes, self.n_base_estimators_])
 
         for i in range(self.n_base_estimators_):

src/autoprognosis/plugins/pipeline/__init__.py

@@ -17,6 +17,7 @@
     _generate_getstate,
     _generate_hyperparameter_space_for_layer_impl,
     _generate_hyperparameter_space_impl,
+    _generate_is_fitted,
     _generate_load,
     _generate_load_template,
     _generate_name_impl,
@@ -37,6 +38,7 @@ def __new__(cls: Type, name: str, plugins: Tuple[Type, ...], dct: dict) -> Any:
         dct["__setstate__"] = _generate_setstate()
         dct["__getstate__"] = _generate_getstate()
         dct["fit"] = _generate_fit()
+        dct["is_fitted"] = _generate_is_fitted()
         dct["predict"] = _generate_predict()
         dct["predict_proba"] = _generate_predict_proba()
         dct["score"] = _generate_score()
@@ -84,6 +86,9 @@ def get_args(*args: Any, **kwargs: Any) -> Dict:
     def fit(self: Any, X: pd.DataFrame, *args: Any, **kwargs: Any) -> Any:
         raise NotImplementedError("not implemented")
 
+    def is_fitted(self: Any) -> Any:
+        raise NotImplementedError("not implemented")
+
     def predict(*args: Any, **kwargs: Any) -> pd.DataFrame:
         raise NotImplementedError("not implemented")
 

src/autoprognosis/plugins/pipeline/generators.py

@@ -128,6 +128,13 @@ def fit_impl(self: Any, X: pd.DataFrame, *args: Any, **kwargs: Any) -> Any:
     return fit_impl
 
 
+def _generate_is_fitted() -> Callable:
+    def fit_impl(self: Any) -> Any:
+        return self.stages[-1].is_fitted()
+
+    return fit_impl
+
+
 def _generate_predict() -> Callable:
     @decorators.benchmark
     def predict_impl(
@@ -249,6 +256,7 @@ def getstate_impl(self: Any) -> dict:
     "_generate_sample_param_impl",
     "_generate_constructor",
     "_generate_fit",
+    "_generate_is_fitted",
     "_generate_predict",
     "_generate_predict_proba",
     "_generate_score",

src/autoprognosis/plugins/prediction/base.py

@@ -43,6 +43,9 @@ def explain(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> pd.DataFrame:
         raise NotImplementedError(f"Explainer not implemented for {self.name()}")
 
     def predict_proba(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> pd.DataFrame:
+        if not self.is_fitted():
+            raise RuntimeError("Fit the model first")
+
         X = self._transform_input(X)
         return pd.DataFrame(self._predict_proba(X, *args, **kwargs))
 

src/autoprognosis/plugins/prediction/classifiers/base.py

@@ -39,7 +39,11 @@ def fit(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> plugin.Plugin:
             raise RuntimeError("Training requires X, y")
         Y = cast.to_dataframe(args[0]).values.ravel()
 
-        return self._fit(X, Y, **kwargs)
+        self._fit(X, Y, **kwargs)
+
+        self._fitted = True
+
+        return self
 
     def score(self, X: pd.DataFrame, y: pd.DataFrame, metric: str = "aucroc") -> float:
         ev = classifier_evaluator(metric)

src/autoprognosis/plugins/prediction/regression/base.py

@@ -43,6 +43,7 @@ def fit(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> "RegressionPlugin":
 
         X = self._fit_input(X)
         self._fit(X, *args, **kwargs)
+        self._fitted = True
 
         return self
 

src/autoprognosis/plugins/prediction/risk_estimation/base.py

@@ -56,6 +56,7 @@ def fit(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> "RiskEstimationPlug
 
         X = self._fit_input(X)
         self._fit(X, *args, **kwargs)
+        self._fitted = True
 
         if self.with_explanations and self.explainer is None:
             if "eval_times" not in kwargs:
@@ -81,6 +82,9 @@ def fit(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> "RiskEstimationPlug
         return self
 
     def explain(self, X: pd.DataFrame, *args: Any, **kwargs: Any) -> pd.DataFrame:
+        if not self.is_fitted():
+            raise RuntimeError("Fit the model first")
+
         X = self._transform_input(X)
         if self.explainer is None:
             raise ValueError("Interpretability is not enabled for this model")

src/autoprognosis/version.py

@@ -1 +1 @@
-__version__ = "0.1.8"
+__version__ = "0.1.9"