Merge pull request #232 from ndem0/v2

Refactoring for pre- and post-processing, new design

Co-authored-by: MMRROOO <mrowkamaad@gmail.com>

ezyrb/__init__.py

@@ -1,23 +1,16 @@
 """EZyRB package"""
 
 __all__ = [
-    'Database', 'Reduction', 'POD', 'Approximation', 'RBF', 'Linear', 'GPR',
+    'Database', 'Snapshot', 'Reduction', 'POD', 'Approximation', 'RBF', 'Linear', 'GPR',
     'ANN', 'KNeighborsRegressor', 'RadiusNeighborsRegressor', 'AE',
     'ReducedOrderModel', 'PODAE', 'RegularGrid'
 ]
 
 from .meta import *
 from .database import Database
-from .reduction import Reduction
-from .pod import POD
-from .ae import AE
-from .pod_ae import PODAE
-from .approximation import Approximation
-from .rbf import RBF
-from .linear import Linear
-from .regular_grid import RegularGrid
-from .gpr import GPR
+from .snapshot import Snapshot
+from .parameter import Parameter
 from .reducedordermodel import ReducedOrderModel
-from .ann import ANN
-from .kneighbors_regressor import KNeighborsRegressor
-from .radius_neighbors_regressor import RadiusNeighborsRegressor
+from .reduction import *
+from .approximation import *
+from .regular_grid import RegularGrid

ezyrb/approximation/__init__.py

@@ -0,0 +1,14 @@
+"""EZyRB package"""
+
+__all__ = [
+    'Approximation', 'RBF', 'Linear', 'GPR',
+    'ANN', 'KNeighborsRegressor', 'RadiusNeighborsRegressor'
+]
+
+from .approximation import Approximation
+from .rbf import RBF
+from .linear import Linear
+from .gpr import GPR
+from .ann import ANN
+from .kneighbors_regressor import KNeighborsRegressor
+from .radius_neighbors_regressor import RadiusNeighborsRegressor

ezyrb/gpr.py → ezyrb/approximation/gpr.py

@@ -16,7 +16,14 @@ class GPR(Approximation):
         arranged by row.
     :cvar numpy.ndarray Y_sample: the array containing the output values,
         arranged by row.
-    :cvar GPy.models.GPRegression model: the regression model.
+    :cvar sklearn.gaussian_process.GaussianProcessRegressor model: the
+        regression model.
+    :cvar sklearn.gaussian_process.kernels.Kernel kern: kernel object from
+        sklearn.
+    :cvar bool normalizer: whether to normilize `values` or not.  Defaults to
+        True.
+    :cvar int optimization_restart: number of restarts for the optimization.
+        Defaults to 20.
 
     :Example:
 
@@ -30,28 +37,21 @@ class GPR(Approximation):
         >>> print(np.allclose(y, y_pred))
 
     """
-    def __init__(self):
+    def __init__(self, kern=None, normalizer=True, optimization_restart=20):
+
         self.X_sample = None
         self.Y_sample = None
+        self.kern = kern
+        self.normalizer = normalizer  # TODO: avoid normalizer inside GPR class
+        self.optimization_restart = optimization_restart
         self.model = None
 
-    def fit(self,
-            points,
-            values,
-            kern=None,
-            normalizer=True,
-            optimization_restart=20):
+    def fit(self, points, values):
         """
         Construct the regression given `points` and `values`.
 
         :param array_like points: the coordinates of the points.
         :param array_like values: the values in the points.
-        :param sklearn.gaussian_process.kernels.Kernel kern: kernel object from
-            sklearn.
-        :param bool normalizer: whether to normilize `values` or not.
-            Defaults to True.
-        :param int optimization_restart: number of restarts for the
-            optimization. Defaults to 20.
         """
         self.X_sample = np.array(points)
         self.Y_sample = np.array(values)
@@ -61,8 +61,8 @@ def fit(self,
             self.Y_sample = self.Y_sample.reshape(-1, 1)
 
         self.model = GaussianProcessRegressor(
-            kernel=kern, n_restarts_optimizer=optimization_restart,
-            normalize_y=normalizer)
+            kernel=self.kern, n_restarts_optimizer=self.optimization_restart,
+            normalize_y=self.normalizer)
         self.model.fit(self.X_sample, self.Y_sample)
 
     def predict(self, new_points, return_variance=False):

ezyrb/linear.py → ezyrb/approximation/linear.py

@@ -41,7 +41,10 @@ def fit(self, points, values):
 
         if as_np_array.ndim == 1 or (as_np_array.ndim == 2
                                      and as_np_array.shape[1] == 1):
-            self.interpolator = interp1d(as_np_array, values, axis=0)
+
+            self.interpolator = interp1d(as_np_array, values, axis=0,
+                                         bounds_error=False,
+                                         fill_value=self.fill_value)
         else:
             self.interpolator = LinearNDInterp(points,
                                                values,

ezyrb/database.py

@@ -2,6 +2,9 @@
 
 import numpy as np
 
+from .parameter import Parameter
+from .snapshot import Snapshot
+
 class Database():
     """
     Database class
@@ -14,66 +17,35 @@ class Database():
         None meaning no scaling.
     :param array_like space: the input spatial data
     """
-    def __init__(self,
-                 parameters=None,
-                 snapshots=None,
-                 scaler_parameters=None,
-                 scaler_snapshots=None,
-                 space=None):
-        self._parameters = None
-        self._snapshots = None
-        self._space = None
-        self.scaler_parameters = scaler_parameters
-        self.scaler_snapshots = scaler_snapshots
-
-        # if only parameters or snapshots are provided
-        if (parameters is None) ^ (snapshots is None):
-            raise RuntimeError(
-                'Parameters and Snapshots are not both provided')
-
-        if space is not None and snapshots is None:
-            raise RuntimeError(
-                'Snapshot data is not provided with Spatial data')
-
-        if parameters is not None and snapshots is not None:
-            if space is not None:
-                self.add(parameters, snapshots, space)
-            else:
-                self.add(parameters, snapshots)
+    def __init__(self, parameters=None, snapshots=None):
+        self._pairs = []
 
+        if parameters is None and snapshots is None:
+            return
 
-    @property
-    def parameters(self):
-        """
-        The matrix containing the input parameters (by row).
-
-        :rtype: numpy.ndarray
-        """
-        if self.scaler_parameters:
-            return self.scaler_parameters.fit_transform(self._parameters)
+        if len(parameters) != len(snapshots):
+            raise ValueError
 
-        return self._parameters
+        for param, snap in zip(parameters, snapshots):
+            self.add(Parameter(param), Snapshot(snap))
 
     @property
-    def snapshots(self):
+    def parameters_matrix(self):
         """
-        The matrix containing the snapshots (by row).
+        The matrix containing the input parameters (by row).
 
         :rtype: numpy.ndarray
         """
-        if self.scaler_snapshots:
-            return self.scaler_snapshots.fit_transform(self._snapshots)
-
-        return self._snapshots
+        return np.asarray([pair[0].values for pair in self._pairs])
 
     @property
-    def space(self):
+    def snapshots_matrix(self):
         """
-        The matrix containing spatial information (by row).
+        The matrix containing the snapshots (by row).
 
         :rtype: numpy.ndarray
         """
-        return self._space
+        return np.asarray([pair[1].flattened for pair in self._pairs])
 
     def __getitem__(self, val):
         """
@@ -83,75 +55,85 @@ def __getitem__(self, val):
         .. warning:: The new parameters and snapshots are a view of the
             original Database.
         """
-        if isinstance(val, int):
-            if self._space is None:
-                return Database(np.reshape(self._parameters[val],
-                                    (1,len(self._parameters[val]))),
-                                np.reshape(self._snapshots[val],
-                                    (1,len(self._snapshots[val]))),
-                                self.scaler_parameters,
-                                self.scaler_snapshots)
-
-            return Database(np.reshape(self._parameters[val],
-                                (1,len(self._parameters[val]))),
-                            np.reshape(self._snapshots[val],
-                                (1,len(self._snapshots[val]))),
-                            self.scaler_parameters,
-                            self.scaler_snapshots,
-                            np.reshape(self._space[val],
-                                (1,len(self._space[val]))))
-
-        if self._space is None:
-            return Database(self._parameters[val],
-                            self._snapshots[val],
-                            self.scaler_parameters,
-                            self.scaler_snapshots)
-
-        return Database(self._parameters[val],
-                        self._snapshots[val],
-                        self.scaler_parameters,
-                        self.scaler_snapshots,
-                        self._space[val])
+        if isinstance(val, np.ndarray):
+            view = Database()
+            for p, s in np.asarray(self._pairs)[val]:
+                view.add(p, s)
+        elif isinstance(val, (int, slice)):
+            view = Database()
+            view._pairs = self._pairs[val]
+        return view
 
     def __len__(self):
         """
         This method returns the number of snapshots.
 
         :rtype: int
         """
-        return len(self._snapshots)
+        return len(self._pairs)
+
+    def __str__(self):
+        """ Print minimal info about the Database """
+        return str(self.parameters_matrix)
 
-    def add(self, parameters, snapshots, space=None):
+    def add(self, parameter, snapshot):
         """
         Add (by row) new sets of snapshots and parameters to the original
         database.
 
-        :param array_like parameters: the parameters to add.
-        :param array_like snapshots: the snapshots to add.
+        :param Parameter parameter: the parameter to add.
+        :param Snapshot snapshot: the snapshot to add.
         """
-        if len(parameters) != len(snapshots):
-            raise RuntimeError(
-                'Different number of parameters and snapshots.')
-
-        if self._space is not None and space is None:
-            raise RuntimeError('No Spatial Value given')
-
-        if (self._space is not None) or (space is not None):
-            if space.shape != snapshots.shape:
-                raise RuntimeError(
-                    'shape of space and snapshots are different.')
-
-        if self._parameters is None and self._snapshots is None:
-            self._parameters = parameters
-            self._snapshots = snapshots
-            if self._space is None:
-                self._space = space
-        elif self._space is None:
-            self._parameters = np.vstack([self._parameters, parameters])
-            self._snapshots = np.vstack([self._snapshots, snapshots])
-        else:
-            self._parameters = np.vstack([self._parameters, parameters])
-            self._snapshots = np.vstack([self._snapshots, snapshots])
-            self._space = np.vstack([self._space, space])
+        if not isinstance(parameter, Parameter):
+            raise ValueError
+
+        if not isinstance(snapshot, Snapshot):
+            raise ValueError
+
+        self._pairs.append((parameter, snapshot))
 
         return self
+
+
+    def split(self, chunks, seed=None):
+        """
+
+        >>> db = Database(...)
+        >>> train, test = db.split([0.8, 0.2]) # ratio
+        >>> train, test = db.split([80, 20])   # n snapshots
+
+        """
+        if all(isinstance(n, int) for n in chunks):
+            if sum(chunks) != len(self):
+                raise ValueError('chunk elements are inconsistent')
+
+            ids = [
+                j for j, chunk in enumerate(chunks)
+                for i in range(chunk)
+            ]
+            np.random.shuffle(ids)
+
+
+        elif all(isinstance(n, float) for n in chunks):
+            if not np.isclose(sum(chunks), 1.):
+                raise ValueError('chunk elements are inconsistent')
+
+            cum_chunks = np.cumsum(chunks)
+            cum_chunks = np.insert(cum_chunks, 0, 0.0)
+            ids = np.ones(len(self)) * -1.
+            tmp = np.random.uniform(0, 1, size=len(self))
+            for i in range(len(cum_chunks)-1):
+                is_between = np.logical_and(
+                    tmp >= cum_chunks[i], tmp < cum_chunks[i+1])
+                ids[is_between] = i
+
+        else:
+            ValueError
+
+        new_database = [Database() for _ in range(len(chunks))]
+        for i, chunk in enumerate(chunks):
+            chunk_ids = np.array(ids) == i
+            for p, s in np.asarray(self._pairs)[chunk_ids]:
+                new_database[i].add(p, s)
+
+        return new_database

ezyrb/parameter.py

@@ -0,0 +1,18 @@
+""" Module for parameter object """
+import numpy as np
+
+class Parameter:
+
+    def __init__(self, values):
+        self.values = values
+
+    @property
+    def values(self):
+        """ Get the snapshot values. """
+        return self._values
+
+    @values.setter
+    def values(self, new_values):
+        if np.asarray(new_values).ndim != 1:
+            raise ValueError('only 1D array are usable as parameter.')
+        self._values = new_values

ezyrb/plugin/__init__.py

@@ -0,0 +1,13 @@
+""" Plugins submodule """
+
+__all__ = [
+    'Plugin',
+    'DatabaseScaler',
+    'ShiftSnapshots',
+    'AutomaticShiftSnapshots',
+]
+
+from .scaler import DatabaseScaler
+from .plugin import Plugin
+from .shift import ShiftSnapshots
+from .automatic_shift import AutomaticShiftSnapshots