Model Builder refactoring

pymc_experimental/model_builder.py

@@ -15,6 +15,7 @@
 
 import hashlib
 import json
+from abc import abstractmethod
 from pathlib import Path
 from typing import Dict, Union
 
@@ -24,12 +25,10 @@
 import pymc as pm
 
 
-class ModelBuilder(pm.Model):
+class ModelBuilder:
     """
     ModelBuilder can be used to provide an easy-to-use API (similar to scikit-learn) for models
     and help with deployment.
-
-    Extends the pymc.Model class.
     """
 
     _model_type = "BaseClass"
@@ -38,8 +37,8 @@ class ModelBuilder(pm.Model):
     def __init__(
         self,
         model_config: Dict,
-        sampler_config: Dict,
-        data: Dict[str, Union[np.ndarray, pd.DataFrame, pd.Series]] = None,
+        data: Dict[str, Union[np.ndarray, pd.DataFrame, pd.Series]],
+        sampler_config: Dict = None,
     ):
         """
         Initializes model configuration and sampler configuration for the model
@@ -48,10 +47,10 @@ def __init__(
         ----------
         model_config : Dictionary
             dictionary of parameters that initialise model configuration. Generated by the user defined create_sample_input method.
-        sampler_config : Dictionary
-            dictionary of parameters that initialise sampler configuration. Generated by the user defined create_sample_input method.
         data : Dictionary
             It is the data we need to train the model on.
+        sampler_config : Dictionary
+            dictionary of parameters that initialise sampler configuration. Generated by the user defined create_sample_input method.
         Examples
         --------
         >>> class LinearModel(ModelBuilder):
@@ -61,19 +60,20 @@ def __init__(
 
         super().__init__()
         self.model_config = model_config  # parameters for priors etc.
-        self.sample_config = sampler_config  # parameters for sampling
-        self.idata = None  # inference data object
+        self.sampler_config = sampler_config  # parameters for sampling
         self.data = data
-        self.build()
+        self.idata = (
+            None  # inference data object placeholder, idata is generated during build execution
+        )
 
-    def build(self):
+    def build(self) -> None:
         """
         Builds the defined model.
         """
 
-        with self:
-            self.build_model(self.model_config, self.data)
+        self.build_model(self, self.model_config, self.data)
 
+    @abstractmethod
     def _data_setter(
         self, data: Dict[str, Union[np.ndarray, pd.DataFrame, pd.Series]], x_only: bool = True
     ):
@@ -100,8 +100,10 @@ def _data_setter(
 
         raise NotImplementedError
 
-    @classmethod
-    def create_sample_input(cls):
+    # need a discussion if it's really needed.
+    @staticmethod
+    @abstractmethod
+    def create_sample_input():
         """
         Needs to be implemented by the user in the inherited class.
         Returns examples for data, model_config, sampler_config.
@@ -135,7 +137,7 @@ def create_sample_input(cls):
 
         raise NotImplementedError
 
-    def save(self, fname):
+    def save(self, fname: str) -> None:
         """
         Saves inference data of the model.
 
@@ -159,8 +161,9 @@ def save(self, fname):
         self.idata.to_netcdf(file)
 
     @classmethod
-    def load(cls, fname):
+    def load(cls, fname: str):
         """
+        Creates a ModelBuilder instance from a file,
         Loads inference data for the model.
 
         Parameters
@@ -170,7 +173,7 @@ def load(cls, fname):
 
         Returns
         -------
-        Returns the inference data that is loaded from local system.
+        Returns an instance of ModelBuilder.
 
         Raises
         ------
@@ -187,22 +190,29 @@ def load(cls, fname):
 
         filepath = Path(str(fname))
         idata = az.from_netcdf(filepath)
-        self = cls(
-            json.loads(idata.attrs["model_config"]),
-            json.loads(idata.attrs["sampler_config"]),
-            idata.fit_data.to_dataframe(),
+        if "sampler_config" in idata.attrs:
+            sampler_config = json.loads(idata.attrs["sampler_config"])
+        else:
+            sampler_config = None
+        model_builder = cls(
+            model_config=json.loads(idata.attrs["model_config"]),
+            sampler_config=sampler_config,
+            data=idata.fit_data.to_dataframe(),
         )
-        self.idata = idata
-        if self.id != idata.attrs["id"]:
+        model_builder.idata = idata
+        model_builder.build()
+        if model_builder.id != idata.attrs["id"]:
             raise ValueError(
-                f"The file '{fname}' does not contain an inference data of the same model or configuration as '{self._model_type}'"
+                f"The file '{fname}' does not contain an inference data of the same model or configuration as '{cls._model_type}'"
             )
 
-        return self
+        return model_builder
 
-    def fit(self, data: Dict[str, Union[np.ndarray, pd.DataFrame, pd.Series]] = None):
+    def fit(
+        self, data: Dict[str, Union[np.ndarray, pd.DataFrame, pd.Series]] = None
+    ) -> az.InferenceData:
         """
-        As the name suggests fit can be used to fit a model using the data that is passed as a parameter.
+        Fit a model using the data passed as a parameter.
         Sets attrs to inference data of the model.
 
         Parameter
@@ -225,35 +235,40 @@ def fit(self, data: Dict[str, Union[np.ndarray, pd.DataFrame, pd.Series]] = None
 
         if data is not None:
             self.data = data
-            self._data_setter(data)
-
-        if self.basic_RVs == []:
             self.build()
+            self._data_setter(data)
 
-        with self:
-            self.idata = pm.sample(**self.sample_config)
+        with self.model:
+            if self.sampler_config is not None:
+                self.idata = pm.sample(**self.sampler_config)
+            else:
+                self.idata = pm.sample()
             self.idata.extend(pm.sample_prior_predictive())
             self.idata.extend(pm.sample_posterior_predictive(self.idata))
 
         self.idata.attrs["id"] = self.id
         self.idata.attrs["model_type"] = self._model_type
         self.idata.attrs["version"] = self.version
-        self.idata.attrs["sampler_config"] = json.dumps(self.sample_config)
+        if self.sampler_config is not None:
+            self.idata.attrs["sampler_config"] = json.dumps(self.sampler_config)
         self.idata.attrs["model_config"] = json.dumps(self.model_config)
         self.idata.add_groups(fit_data=self.data.to_xarray())
         return self.idata
 
     def predict(
         self,
         data_prediction: Dict[str, Union[np.ndarray, pd.DataFrame, pd.Series]] = None,
-    ):
+        extend_idata: bool = True,
+    ) -> dict:
         """
         Uses model to predict on unseen data and return point prediction of all the samples
 
         Parameters
         ---------
         data_prediction : Dictionary of string and either of numpy array, pandas dataframe or pandas Series
             It is the data we need to make prediction on using the model.
+        extend_idata : Boolean determining whether the predictions should be added to inference data object.
+            Defaults to True.
 
         Returns
         -------
@@ -275,7 +290,8 @@ def predict(
 
         with self.model:  # sample with new input data
             post_pred = pm.sample_posterior_predictive(self.idata)
-
+            if extend_idata:
+                self.idata.extend(post_pred)
         # reshape output
         post_pred = self._extract_samples(post_pred)
         for key in post_pred:
@@ -286,16 +302,17 @@ def predict(
     def predict_posterior(
         self,
         data_prediction: Dict[str, Union[np.ndarray, pd.DataFrame, pd.Series]] = None,
-    ):
+        extend_idata: bool = True,
+    ) -> Dict[str, np.array]:
         """
         Uses model to predict samples on unseen data.
 
         Parameters
         ---------
         data_prediction : Dictionary of string and either of numpy array, pandas dataframe or pandas Series
             It is the data we need to make prediction on using the model.
-        point_estimate : bool
-            Adds point like estimate used as mean passed as
+        extend_idata : Boolean determining whether the predictions should be added to inference data object.
+            Defaults to True.
 
         Returns
         -------
@@ -317,6 +334,8 @@ def predict_posterior(
 
         with self.model:  # sample with new input data
             post_pred = pm.sample_posterior_predictive(self.idata)
+            if extend_idata:
+                self.idata.extend(post_pred)
 
         # reshape output
         post_pred = self._extract_samples(post_pred)
@@ -357,5 +376,4 @@ def id(self) -> str:
         hasher.update(str(self.model_config.values()).encode())
         hasher.update(self.version.encode())
         hasher.update(self._model_type.encode())
-        # hasher.update(str(self.sample_config.values()).encode())
         return hasher.hexdigest()[:16]

pymc_experimental/tests/test_model_builder.py

@@ -12,7 +12,6 @@
 #   See the License for the specific language governing permissions and
 #   limitations under the License.
 
-
 import hashlib
 import sys
 import tempfile
@@ -29,26 +28,39 @@ class test_ModelBuilder(ModelBuilder):
     _model_type = "LinearModel"
     version = "0.1"
 
-    def build_model(self, model_config, data=None):
-        if data is not None:
-            x = pm.MutableData("x", data["input"].values)
-            y_data = pm.MutableData("y_data", data["output"].values)
-
-        # prior parameters
-        a_loc = model_config["a_loc"]
-        a_scale = model_config["a_scale"]
-        b_loc = model_config["b_loc"]
-        b_scale = model_config["b_scale"]
-        obs_error = model_config["obs_error"]
-
-        # priors
-        a = pm.Normal("a", a_loc, sigma=a_scale)
-        b = pm.Normal("b", b_loc, sigma=b_scale)
-        obs_error = pm.HalfNormal("σ_model_fmc", obs_error)
-
-        # observed data
-        if data is not None:
-            y_model = pm.Normal("y_model", a + b * x, obs_error, shape=x.shape, observed=y_data)
+    def build_model(
+        self,
+        model_instance: ModelBuilder,
+        model_config: dict,
+        data: dict = None,
+        sampler_config: dict = None,
+    ):
+        model_instance.model_config = model_config
+        model_instance.data = data
+        self.model_config = model_config
+        self.sampler_config = sampler_config
+        self.data = data
+
+        with pm.Model() as model_instance.model:
+            if data is not None:
+                x = pm.MutableData("x", data["input"].values)
+                y_data = pm.MutableData("y_data", data["output"].values)
+
+            # prior parameters
+            a_loc = model_config["a_loc"]
+            a_scale = model_config["a_scale"]
+            b_loc = model_config["b_loc"]
+            b_scale = model_config["b_scale"]
+            obs_error = model_config["obs_error"]
+
+            # priors
+            a = pm.Normal("a", a_loc, sigma=a_scale)
+            b = pm.Normal("b", b_loc, sigma=b_scale)
+            obs_error = pm.HalfNormal("σ_model_fmc", obs_error)
+
+            # observed data
+            if data is not None:
+                y_model = pm.Normal("y_model", a + b * x, obs_error, shape=x.shape, observed=y_data)
 
     def _data_setter(self, data: pd.DataFrame):
         with self.model:
@@ -57,7 +69,7 @@ def _data_setter(self, data: pd.DataFrame):
                 pm.set_data({"y_data": data["output"].values})
 
     @classmethod
-    def create_sample_input(cls):
+    def create_sample_input(self):
         x = np.linspace(start=0, stop=1, num=100)
         y = 5 * x + 3
         y = y + np.random.normal(0, 1, len(x))
@@ -81,14 +93,14 @@ def create_sample_input(cls):
         return data, model_config, sampler_config
 
     @staticmethod
-    def initial_build_and_fit(check_idata=True):
+    def initial_build_and_fit(check_idata=True) -> ModelBuilder:
         data, model_config, sampler_config = test_ModelBuilder.create_sample_input()
-        model = test_ModelBuilder(model_config, sampler_config, data)
-        model.fit()
+        model_builder = test_ModelBuilder(model_config, sampler_config, data)
+        model_builder.idata = model_builder.fit(data=data)
         if check_idata:
-            assert model.idata is not None
-            assert "posterior" in model.idata.groups()
-        return model
+            assert model_builder.idata is not None
+            assert "posterior" in model_builder.idata.groups()
+        return model_builder
 
 
 def test_fit():
@@ -105,16 +117,16 @@ def test_fit():
     sys.platform == "win32", reason="Permissions for temp files not granted on windows CI."
 )
 def test_save_load():
-    model = test_ModelBuilder.initial_build_and_fit(False)
+    test_builder = test_ModelBuilder.initial_build_and_fit()
     temp = tempfile.NamedTemporaryFile(mode="w", encoding="utf-8", delete=False)
-    model.save(temp.name)
-    model2 = test_ModelBuilder.load(temp.name)
-    assert model.idata.groups() == model2.idata.groups()
+    test_builder.save(temp.name)
+    test_builder2 = test_ModelBuilder.load(temp.name)
+    assert test_builder.idata.groups() == test_builder2.idata.groups()
 
     x_pred = np.random.uniform(low=0, high=1, size=100)
     prediction_data = pd.DataFrame({"input": x_pred})
-    pred1 = model.predict(prediction_data)
-    pred2 = model2.predict(prediction_data)
+    pred1 = test_builder.predict(prediction_data)
+    pred2 = test_builder2.predict(prediction_data)
     assert pred1["y_model"].shape == pred2["y_model"].shape
     temp.close()