Inverted steps 2 and 3

tournesol-app · Dec 20, 2024 · 28d5ca0 · 28d5ca0
1 parent 968e4fc
commit 28d5ca0
Show file tree

Hide file tree

Showing 4 changed files with 32 additions and 25 deletions.
diff --git a/solidago/src/solidago/pipeline/pipeline.py b/solidago/src/solidago/pipeline/pipeline.py
@@ -10,8 +10,8 @@
 from solidago.scoring_model import ScoringModel, ScaledScoringModel
 
 from solidago.trust_propagation import TrustPropagation, TrustAll, LipschiTrust, NoTrustPropagation
-from solidago.voting_rights import VotingRights, VotingRightsAssignment, AffineOvertrust, IsTrust
 from solidago.preference_learning import PreferenceLearning, UniformGBT
+from solidago.voting_rights import VotingRights, VotingRightsAssignment, AffineOvertrust, IsTrust
 from solidago.scaling import Scaling, ScalingCompose, Mehestan, QuantileZeroShift, Standardize, NoScaling
 from solidago.aggregation import Aggregation, StandardizedQrMedian, StandardizedQrQuantile, Average, EntitywiseQrQuantile
 from solidago.post_process import PostProcess, Squash, NoPostProcess
@@ -33,16 +33,16 @@ class DefaultPipeline:
         sink_vouch=5.0,
         error=1e-8
     )
-    voting_rights: VotingRightsAssignment = AffineOvertrust(
-        privacy_penalty=0.5, 
-        min_overtrust=2.0,
-        overtrust_ratio=0.1,
-    )
     preference_learning: PreferenceLearning = UniformGBT(
         prior_std_dev=7,
         convergence_error=1e-5,
         cumulant_generating_function_error=1e-5,
     )
+    voting_rights: VotingRightsAssignment = AffineOvertrust(
+        privacy_penalty=0.5, 
+        min_overtrust=2.0,
+        overtrust_ratio=0.1,
+    )
     scaling: Scaling = ScalingCompose(
         Mehestan(
             lipschitz=0.1,
@@ -77,8 +77,8 @@ class Pipeline:
     def __init__(
         self,
         trust_propagation: TrustPropagation = DefaultPipeline.trust_propagation,
-        voting_rights: VotingRightsAssignment = DefaultPipeline.voting_rights,
         preference_learning: PreferenceLearning = DefaultPipeline.preference_learning,
+        voting_rights: VotingRightsAssignment = DefaultPipeline.voting_rights,
         scaling: Scaling = DefaultPipeline.scaling,
         aggregation: Aggregation = DefaultPipeline.aggregation,
         post_process: PostProcess = DefaultPipeline.post_process,
@@ -102,8 +102,8 @@ def __init__(
             and make it readily usable for applications.
         """
         self.trust_propagation = trust_propagation
-        self.voting_rights = voting_rights
         self.preference_learning = preference_learning
+        self.voting_rights = voting_rights
         self.scaling = scaling
         self.aggregation = aggregation
         self.post_process = post_process
@@ -113,8 +113,8 @@ def __init__(
     def from_json(cls, json) -> "Pipeline":
         return Pipeline(
             trust_propagation=trust_propagation_from_json(json["trust_propagation"]),
-            voting_rights=voting_rights_from_json(json["voting_rights"]),
             preference_learning=preference_learning_from_json(json["preference_learning"]),
+            voting_rights=voting_rights_from_json(json["voting_rights"]),
             scaling=scaling_from_json(json["scaling"]),
             aggregation=aggregation_from_json(json["aggregation"]),
             post_process=post_process_from_json(json["post_process"]),
@@ -185,20 +185,20 @@ def __call__(
         logger.info(f"Pipeline 1. Terminated in {np.round(start_step2 - start_step1, 2)} seconds")
         if output is not None:
             output.save_trust_scores(trusts=users)
+
+        logger.info(f"Pipeline 2. Learning preferences with {str(self.preference_learning)}")
+        user_models = self.preference_learning(judgments, users, entities, init_user_models)
+        start_step3 = timeit.default_timer()
+        logger.info(f"Pipeline 2. Terminated in {np.round(start_step3 - start_step2, 2)} seconds")
+        raw_scorings = user_models
 
-        logger.info(f"Pipeline 2. Computing voting rights with {str(self.voting_rights)}")
+        logger.info(f"Pipeline 3. Computing voting rights with {str(self.voting_rights)}")
         # WARNING: `privacy` may contain (user, entity) even if user has expressed no judgement
         # about the entity. These users should not be given a voting right on the entity.
         # For now, irrelevant privacy values are excluded in `input.get_pipeline_kwargs()`
-        voting_rights, entities = self.voting_rights(users, entities, vouches, privacy)
-        start_step3 = timeit.default_timer()
-        logger.info(f"Pipeline 2. Terminated in {np.round(start_step3 - start_step2, 2)} seconds")
-
-        logger.info(f"Pipeline 3. Learning preferences with {str(self.preference_learning)}")
-        user_models = self.preference_learning(judgments, users, entities, init_user_models)
+        voting_rights, entities = self.voting_rights(users, entities, vouches, privacy, user_models)
         start_step4 = timeit.default_timer()
-        logger.info(f"Pipeline 3. Terminated in {int(start_step4 - start_step3)} seconds")
-        raw_scorings = user_models
+        logger.info(f"Pipeline 3. Terminated in {np.round(start_step4 - start_step3, 2)} seconds")
 
         logger.info(f"Pipeline 4. Collaborative scaling with {str(self.scaling)}")
         user_models = self.scaling(user_models, users, entities, voting_rights, privacy)

diff --git a/solidago/src/solidago/voting_rights/affine_overtrust.py b/solidago/src/solidago/voting_rights/affine_overtrust.py
@@ -1,3 +1,5 @@
+from typing import Optional
+
 import numpy as np
 import pandas as pd
 
@@ -30,6 +32,7 @@ def __call__(
         entities: pd.DataFrame,
         vouches: pd.DataFrame,
         privacy: PrivacySettings,
+        user_models: Optional[dict[int, "ScoringModel"]]=None
     ) -> tuple[VotingRights, pd.DataFrame]:
         """Compute voting rights
 
@@ -40,11 +43,11 @@ def __call__(
             * trust_score (float)
         entities: DataFrame with columns
             * entity_id (int, index)
-        vouches: DataFrame
-            This is not used by VotingRightsWithLimitedOvertrust
+        vouches: not used
         privacy: PrivacySettings
             privacy[user, entity] is the privacy setting of user for entity
             May be True, False or None
+        user_models: not used
 
         Returns
         -------

diff --git a/solidago/src/solidago/voting_rights/base.py b/solidago/src/solidago/voting_rights/base.py
@@ -15,6 +15,7 @@ def __call__(
         entities: pd.DataFrame,
         vouches: pd.DataFrame,
         privacy: PrivacySettings,
+        user_models: dict[int, "ScoringModel"]
     ) -> tuple[VotingRights, pd.DataFrame]:
         """ Compute voting rights
         
@@ -31,6 +32,8 @@ def __call__(
             * vouch (float)
         privacy: PrivacySettings
             privacy[user, entity] in { True, False, None }
+        user_models: dict[int, ScoringModel]
+            user_models[user_id] is the user's scoring model
         
         Returns
         -------

diff --git a/solidago/src/solidago/voting_rights/is_trust.py b/solidago/src/solidago/voting_rights/is_trust.py
@@ -1,3 +1,5 @@
+from typing import Optional
+
 import pandas as pd
 import numpy as np
 
@@ -14,7 +16,8 @@ def __call__(
         users: pd.DataFrame,
         entities: pd.DataFrame,
         vouches: pd.DataFrame,
-        privacy: pd.DataFrame
+        privacy: "PrivacySettings",
+        user_models: Optional[dict[int, "ScoringModel"]]=None
     ) -> tuple[VotingRights, pd.DataFrame]:
         """ Compute voting rights
         
@@ -25,12 +28,10 @@ def __call__(
             * trust_score (float)
         entities: DataFrame with columns
             * entity_id (int, index)
-        vouches: DataFrame with columns
-            * voucher (int)
-            * vouchee (int)
-            * vouch (float)
+        vouches: Not used
         privacy: PrivacySettings
             privacy[user, entity] in { True, False, None }
+        user_models: Not used
         
         Returns
         -------