approval validation

docs-source/source/validation/approval.rst

@@ -2,3 +2,49 @@ Approval Samplers
 =================
 
 We present below the validation tests we ran for approval samplers.
+
+Impartial
+---------
+
+:py:func:`prefsampling.approval.impartial`
+
+The probability distribution generated by the approval impartial culture model.
+
+.. image:: validation_plots/approval/impartial.png
+  :width: 800
+  :alt: Observed versus theoretical frequencies for the impartial culture
+
+Resampling
+----------
+
+:py:func:`prefsampling.approval.resampling`
+
+The probability distribution generated by the resampling model.
+See `Szufa et al. (2022) <https://www.ijcai.org/proceedings/2022/0071.pdf>`_
+for more details.
+
+.. image:: validation_plots/approval/resampling.png
+  :width: 800
+  :alt: Observed versus theoretical frequencies for the resampling model
+
+Identity
+----------
+
+:py:func:`prefsampling.approval.identity`
+
+The probability distribution generated by the identity model.
+
+.. image:: validation_plots/approval/identity.png
+  :width: 800
+  :alt: Observed versus theoretical frequencies for the identity model
+
+Noise
+----------
+
+:py:func:`prefsampling.approval.noise`
+
+The probability distribution generated by the noise model based on the Hamming distance.
+
+.. image:: validation_plots/approval/noise.png
+  :width: 800
+  :alt: Observed versus theoretical frequencies for the noise model

prefsampling/approval/noise.py

@@ -79,7 +79,7 @@ def noise(
 
     rng = np.random.default_rng(seed)
 
-    k = int(p * num_candidates)
+    k = math.floor(p * num_candidates)
 
     A = {i for i in range(k)}
     B = set(range(num_candidates)) - A

prefsampling/approval/resampling.py

@@ -1,7 +1,9 @@
 import copy
 
+import math
 import numpy as np
 
+
 from prefsampling.inputvalidators import validate_num_voters_candidates
 
 
@@ -52,7 +54,7 @@ def resampling(
 
     rng = np.random.default_rng(seed)
 
-    k = int(p * num_candidates)
+    k = math.floor(p * num_candidates)
     if central_vote is None:
         central_vote = set(range(k))
 
@@ -122,7 +124,7 @@ def disjoint_resampling(
     rng = np.random.default_rng(seed)
 
     num_groups = g
-    k = int(p * num_candidates)
+    k = math.floor(p * num_candidates)
 
     votes = [set() for _ in range(num_voters)]
 
@@ -189,7 +191,7 @@ def moving_resampling(
 
     breaks = [int(num_voters / num_legs) * i for i in range(num_legs)]
 
-    k = int(p * num_candidates)
+    k = math.floor(p * num_candidates)
     central_vote = {i for i in range(k)}
     ccc = copy.deepcopy(central_vote)
 

validation/approval/impartial.py

@@ -17,7 +17,7 @@ def __init__(self):
         super(ApprovalImpartialValidator, self).__init__(
             parameters_list,
             "Impartial",
-            "impartial_apr",
+            "impartial",
             True,
             sampler_func=impartial,
             constant_parameters="num_voters",

validation/approval/noise.py

@@ -8,12 +8,14 @@
 class ApprovalNoiseValidator(Validator):
     def __init__(self):
         parameters_list = [
-            {"num_voters": 1, "num_candidates": 4, "phi": 0.25, "p": 0.25},
-            {"num_voters": 1, "num_candidates": 4, "phi": 0.25, "p": 0.5},
-            {"num_voters": 1, "num_candidates": 4, "phi": 0.25, "p": 0.75},
-            {"num_voters": 1, "num_candidates": 4, "phi": 0.5, "p": 0.25},
-            {"num_voters": 1, "num_candidates": 4, "phi": 0.5, "p": 0.5},
-            {"num_voters": 1, "num_candidates": 4, "phi": 0.5, "p": 0.75},
+            {"num_voters": 1, "num_candidates": 6, "phi": 0.25, "p": 0.5},
+            {"num_voters": 1, "num_candidates": 6, "phi": 0.5, "p": 0.5},
+            {"num_voters": 1, "num_candidates": 6, "phi": 0.75, "p": 0.5},
+            {"num_voters": 1, "num_candidates": 6, "phi": 1., "p": 0.5},
+            {"num_voters": 1, "num_candidates": 6, "phi": 0.25, "p": 0.34},
+            {"num_voters": 1, "num_candidates": 6, "phi": 0.5, "p": 0.34},
+            {"num_voters": 1, "num_candidates": 6, "phi": 0.75, "p": 0.34},
+            {"num_voters": 1, "num_candidates": 6, "phi": 1., "p": 0.34},
         ]
         super(ApprovalNoiseValidator, self).__init__(
             parameters_list,

validation/approval/resampling.py

@@ -11,9 +11,11 @@ def __init__(self):
             {"num_voters": 1, "num_candidates": 6, "phi": 0.25, "p": 0.5},
             {"num_voters": 1, "num_candidates": 6, "phi": 0.5, "p": 0.5},
             {"num_voters": 1, "num_candidates": 6, "phi": 0.75, "p": 0.5},
+            {"num_voters": 1, "num_candidates": 6, "phi": 1., "p": 0.5},
             {"num_voters": 1, "num_candidates": 6, "phi": 0.25, "p": 0.34},
             {"num_voters": 1, "num_candidates": 6, "phi": 0.5, "p": 0.34},
             {"num_voters": 1, "num_candidates": 6, "phi": 0.75, "p": 0.34},
+            {"num_voters": 1, "num_candidates": 6, "phi": 1., "p": 0.34},
         ]
         super(ApprovalResamplingValidator, self).__init__(
             parameters_list,
@@ -53,3 +55,49 @@ def theoretical_distribution(self, sampler_parameters, all_outcomes) -> dict:
     def sample_cast(self, sample):
         return str(sample[0])
 
+
+class ApprovalDisjointResamplingValidator(Validator):
+    def __init__(self):
+        parameters_list = [
+            {"num_voters": 1, "num_candidates": 4, "phi": 0.5, "p": 0.25, "g": 2},
+            # {"num_voters": 1, "num_candidates": 6, "phi": 0.5, "p": 0.33, "g": 2},
+            # {"num_voters": 1, "num_candidates": 6, "phi": 0.5, "p": 0.25, "g": 3},
+            # {"num_voters": 1, "num_candidates": 6, "phi": 0.5, "p": 0.33, "g": 3},
+        ]
+        super(ApprovalDisjointResamplingValidator, self).__init__(
+            parameters_list,
+            "Disjoint Resampling",
+            "disjoint_resampling",
+            True,
+            sampler_func=resampling,
+            constant_parameters=("num_voters", "num_candidates", "phi"),
+            faceted_parameters=("phi", "p"),
+        )
+
+    def all_outcomes(self, sampler_parameters):
+        return get_all_subsets(sampler_parameters["num_candidates"])
+
+    def theoretical_distribution(self, sampler_parameters, all_outcomes) -> dict:
+        m = sampler_parameters["num_candidates"]
+        p = sampler_parameters["p"]
+        phi = sampler_parameters["phi"]
+        k = math.floor(p*m)
+        central_vote = {i for i in range(k)}
+
+        A = {}
+        for outcome in all_outcomes:
+            prob = 1
+            for c in range(m):
+                if c in central_vote and c in outcome:
+                    prob *= (1-phi) + phi*p
+                elif c in central_vote and c not in outcome:
+                    prob *= phi*(1-p)
+                elif c not in central_vote and c in outcome:
+                    prob *= phi*p
+                else:
+                    prob *= (1-phi) + phi*(1-p)
+            A[str(outcome)] = prob
+        return A
+
+    def sample_cast(self, sample):
+        return str(sample[0])

validation/run.py

@@ -4,7 +4,10 @@
 from validation.approval.impartial import ApprovalImpartialValidator
 from validation.approval.identity import ApprovalIdentityValidator
 from validation.approval.noise import ApprovalNoiseValidator
-from validation.approval.resampling import ApprovalResamplingValidator
+from validation.approval.resampling import (
+    ApprovalResamplingValidator,
+    ApprovalDisjointResamplingValidator,
+)
 
 from prefsampling.core.euclidean import EuclideanSpace
 from prefsampling.ordinal import TreeSampler
@@ -47,8 +50,9 @@
 ALL_APPROVAL_VALIDATORS = [
     # ApprovalImpartialValidator(),
     # ApprovalIdentityValidator(),
-    # ApprovalNoiseValidator(),
-    ApprovalResamplingValidator(),
+    ApprovalNoiseValidator(),
+    # ApprovalResamplingValidator(),
+    # ApprovalDisjointResamplingValidator(),
 ]
 
 ALL_ORDINAL_VALIDATORS = [
@@ -79,7 +83,7 @@
 if __name__ == "__main__":
     logging.basicConfig(format="%(levelname)s: %(message)s", level=logging.INFO)
 
-    num_observations = 10000
+    num_observations = 1000
 
     for validator_list, nickname in [
         (ALL_APPROVAL_VALIDATORS, "approval"),