Added initial commit UCB learner

src/lop/active_learning/UCBLearner.py

@@ -0,0 +1,90 @@
+# Copyright 2023 Ian Rankin
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy of this
+# software and associated documentation files (the "Software"), to deal in the Software
+# without restriction, including without limitation the rights to use, copy, modify, merge,
+# publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons
+# to whom the Software is furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all copies or
+# substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
+# INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR
+# PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE
+# FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+# OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+# DEALINGS IN THE SOFTWARE.
+
+# UCBLearner.py
+# Written Ian Rankin - December 2023
+#
+# Upper confidence bound learning algorithm
+
+import numpy as np
+
+from lop.active_learning import ActiveLearner
+from lop.models import PreferenceGP, GP, PreferenceLinear
+
+class UCBLearner(ActiveLearner):
+    ## Constructor
+    # @param alpha - the scaler value on the UCB equation UCB = mean + alpha*sqrt(variance)
+    def __init__(self, alpha=1):
+        super(UCBLearner, self).__init__()
+        self.alpha = alpha
+
+    ## select
+    # Selects the given points
+    # @param candidate_pts - a numpy array of points (nxk), n = number points, k = number of dimmensions
+    # @param num_alts - the number of alterantives to selec (including the highest mean)
+    # @param prev_selection - [opt, default = []]a list of indicies that 
+    # @param prefer_num - [default = None] the points at the start of the candidates
+    #                   to prefer selecting from. Returned as:
+    #                   a. A number of points at the start of canididate_pts to prefer
+    #                   b. A set of points to prefer to select.
+    #                   c. 'pareto' to indicate 
+    #                   d. Enter 0 explicitly ignore selections
+    #                   e. None (default) assumes 0 unless default to pareto is true.
+    # @param return_not - [opt default-false] returns the not selected points when there
+    #                   a preference to selecting to certian points. [] if not but set to true.
+    #                   
+    #
+    # @return [highest_mean, highest_selection, next highest selection, ...],
+    #          selection values for candidate_pts,
+    #          only returns highest mean if "always select best is set"
+    def select(self, candidate_pts, num_alts, prev_selection=[], prefer_pts=None, not_selected=False):
+        prefer_pts = self.get_prefered_set_of_pts(candidate_pts, prefer_pts)
+
+        if isinstance(self.model, (PreferenceGP, GP)):
+            mu, variance = self.gp.predict(candidate_pts)
+            UCB = mu + self.alpha*np.sqrt(variance)
+        elif isinstance(self.model, PreferenceLinear):
+            UCB = 1 # TODO
+        else:
+            raise Exception("UCBLearner does not know how to handle model of type: " + str(type(self.model)))
+
+        best_idx = np.argmax(mu)
+
+        selected_idx = self.select_best_k(UCB, num_alts, best_idx, prefer_num)
+
+        return selected_idx, UCB[selected_idx], mu[best_idx]
+
+
+
+    def select_greedy(self, cur_selection, data):
+        mu, variance, cov, prefer_num = data
+
+        best_v = -float('inf')
+        best_i = -1
+
+        exp_v = 1.0 / (len(cur_selection) + 1)
+        for i in [x for x in range(len(mu)) if x not in cur_selection]:
+            vari = variance[i]
+
+            value = (1-self.alpha)*mu[i] + self.alpha*np.sqrt(vari)
+
+            if value > best_v:
+                best_v = value
+                best_i = i
+
+        return best_i, best_v