Merge pull request #1156 from AMS-Hippo/master

Small default functions for landmark-UMAP

umap/parametric_umap.py

@@ -114,6 +114,8 @@ def __init__(
         self.global_correlation_loss_weight = global_correlation_loss_weight
         self.landmark_loss_fn = landmark_loss_fn
         self.landmark_loss_weight = landmark_loss_weight
+        self.prev_epoch_X = None
+        self.window_vals = None
 
         self.reconstruction_validation = (
             reconstruction_validation  # holdout data for reconstruction acc
@@ -174,6 +176,16 @@ def fit(self, X, y=None, precomputed_distances=None, landmark_positions=None):
             The desired position in low-dimensional space of each sample in X.
             Points that are not landmarks should have nan coordinates.
         """
+        if (self.prev_epoch_X is not None) & (landmark_positions is None):
+            # Add the landmark points for training, then make a landmark vector.
+            landmark_positions = np.stack(
+                [np.array([np.nan, np.nan])]*X.shape[0] + list(
+                    self.transform(
+                        self.prev_epoch_X
+                    )
+                )
+            )
+            X = np.concatenate((X, self.prev_epoch_X))
 
         if landmark_positions is not None:
             len_X = len(X)
@@ -230,6 +242,16 @@ def fit_transform(
             The desired position in low-dimensional space of each sample in X.
             Points that are not landmarks should have nan coordinates.
         """
+        if (self.prev_epoch_X is not None) & (landmark_positions is None):
+            # Add the landmark points for training, then make a landmark vector.
+            landmark_positions = np.stack(
+                [np.array([np.nan, np.nan])]*X.shape[0] + list(
+                    self.transform(
+                        self.prev_epoch_X
+                    )
+                )
+            )
+            X = np.concatenate((X, self.prev_epoch_X))
 
         if landmark_positions is not None:
             len_X = len(X)
@@ -473,6 +495,56 @@ def save(self, save_location, verbose=True):
             if verbose:
                 print("Pickle of ParametricUMAP model saved to {}".format(model_output))
 
+    def add_landmarks(
+        self,
+        X,
+        sample_pct=0.01,
+        sample_mode="uniform",
+        landmark_loss_weight=0.01,
+        idx=None,
+    ):
+        """Add some points from a dataset X as "landmarks."
+
+        Parameters
+        ----------
+        X : array, shape (n_samples, n_features)
+            Old data to be retained.
+        sample_pct : float, optional
+            Percentage of old data to use as landmarks.
+        sample_mode : str, optional
+            Method for sampling points. Allows "uniform" and "predefined."
+        landmark_loss_weight : float, optional
+            Multiplier for landmark loss function.
+
+        """
+        self.sample_pct = sample_pct
+        self.sample_mode = sample_mode
+        self.landmark_loss_weight = landmark_loss_weight
+
+        if self.sample_mode == "uniform":
+            self.prev_epoch_idx = list(
+                np.random.choice(
+                    range(X.shape[0]), int(X.shape[0]*sample_pct), replace=False
+                )
+            )
+            self.prev_epoch_X = X[self.prev_epoch_idx]
+        elif self.sample_mode == "predetermined":
+            if idx is None:
+                raise ValueError(
+                    "Choice of sample_mode is not supported."
+                )
+            else:
+                self.prev_epoch_idx = idx
+                self.prev_epoch_X = X[self.prev_epoch_idx]
+
+        else:
+            raise ValueError(
+                "Choice of sample_mode is not supported."
+            )
+
+    def remove_landmarks(self):
+        self.prev_epoch_X = None
+
     def to_ONNX(self, save_location):
         """Exports trained parametric UMAP as ONNX."""
         # Extract encoder