desdeo 2.0.0__py3-none-any.whl → 2.1.0__py3-none-any.whl

desdeo/api/routers/gdm/gdm_score_bands/gdm_score_bands_manager.py

@@ -0,0 +1,398 @@
+"""GDM Score Bands manager implementation."""
+
+import copy
+
+import polars as pl
+from sqlmodel import Session, select
+
+from desdeo.api.models import (
+    GDMSCOREBandFinalSelection,
+    GDMSCOREBandInformation,
+    Group,
+    GroupIteration,
+    ProblemDB,
+    User,
+)
+from desdeo.api.routers.gdm.gdm_base import GroupManager, ManagerError
+from desdeo.gdm.score_bands import SCOREBandsGDMConfig, SCOREBandsGDMResult, score_bands_gdm
+from desdeo.gdm.voting_rules import consensus_rule, majority_rule
+from desdeo.tools.score_bands import score_json
+
+
+class GDMScoreBandsManager(GroupManager):
+    """The group manager implementation for GDM Score Bands."""
+
+    def __init__(self, group_id: int, db_session: Session):
+        """Initialize the group manager.
+
+        Args:
+            group_id (int): id of the group of this manager.
+            db_session (Session): database session from which the discrete representation of a problem is fetched.
+        """
+        super().__init__(group_id, db_session)
+        # LOAD THE DISCRETE REPRESENTATION
+        group: Group = db_session.exec(select(Group).where(Group.id == group_id)).first()
+        problem: ProblemDB = db_session.exec(select(ProblemDB).where(ProblemDB.id == group.problem_id)).first()
+        if problem.discrete_representation is None:
+            raise ManagerError("The group's discrete representation does not exist!")
+        self.discrete_representation = problem.discrete_representation
+        db_session.close()
+
+    async def run_method(self, user_id: int, data: str):
+        """Method runner implementation for GDM Score Bands.
+
+        Args:
+            user_id (int): The user's ID this data is from.
+            data (str): The data itself. To be validated.
+            WE MIGHT NOT EVEN NEED THIS!!
+        """
+        async with self.lock:
+            await self.send_message(
+                "THIS METHOD IS USED THROUGH THE APPROPRIATE HTTP ENDPOINTS!", self.sockets[user_id]
+            )
+
+    async def vote(self, user: User, group: Group, voted_index: int, session: Session):
+        """A method for voting on a specific band.
+
+        Use this from an actual HTTP endpoint and not in a
+        stupid websocket way like in GNIMBUS. Every time this is
+        operated, send all connected websockets info on the voting.
+
+        Args:
+            user (User): The user that sent the vote
+            group (Group): The group the user belongs to and generally the one we're dealing with
+            voted_index (int): the vote
+            session: (Session) the database session.
+        """
+        async with self.lock:  # not sure if async lock is all that necessary but here we go anyways.
+            group_iteration = session.exec(
+                select(GroupIteration).where(GroupIteration.id == group.head_iteration_id)
+            ).first()
+            if group_iteration is None:
+                raise ManagerError("No such Group Iteration! Did you initialize this group?")
+            info_container = copy.deepcopy(group_iteration.info_container)
+            # if user.id in info_container.user_confirms:
+            #    raise ManagerError("User has already confirmed they want to move on!")
+            info_container.user_votes[str(user.id)] = voted_index
+            group_iteration.info_container = info_container
+            session.add(group_iteration)
+            session.commit()
+            session.refresh(group_iteration)
+
+            # Then update preferences dictionary.
+            await self.broadcast("UPDATE: A vote has been cast.")
+
+    async def confirm(self, user: User, group: Group, session: Session):
+        """A method for a user to indicate that we could move forward with clustering anew.
+
+        After everyone has hit this endpoint, do the following shenanigans:
+            1. Filter the active solution indices using the voting result.
+            2. Create a new GroupIteration.
+            3. Re-cluster the active solutions and put the result into the
+               info_container item of the newly created GroupIteration.
+            4. Send all connected websockets info that we've got some hot
+               new data to update the UI with. Then use some other endpoint to
+               fetch that data.
+
+        Args:
+            user (User): The user
+            group (Group): The group
+            session (Session): The database session.
+        """
+        async with self.lock:
+            if user.id not in group.user_ids:
+                raise ManagerError(
+                    detail=f"User with ID {user.id} is not part of group with ID {group.id}",
+                )
+            group_iteration = session.exec(
+                select(GroupIteration).where(GroupIteration.id == group.head_iteration_id)
+            ).first()
+            if group_iteration is None:
+                raise ManagerError("No group iterations! Did you initialize this group?")
+
+            info_container = copy.deepcopy(group_iteration.info_container)
+            if str(user.id) not in info_container.user_votes:
+                raise ManagerError("User hasn't voted! Cannot confirm!")
+                # if user.id in info_container.user_confirms:
+                raise ManagerError("User has already confirmed they want to move on!")
+            if user.id not in info_container.user_confirms:
+                info_container.user_confirms.append(user.id)
+            group_iteration.info_container = info_container
+            session.add(group_iteration)
+            session.commit()
+            session.refresh(group_iteration)
+
+            # After everyone's done, filter etc.
+            for uid in group.user_ids:
+                # Check if user is not in the list of confirms
+                if uid not in info_container.user_confirms:
+                    return
+
+            # We're in consensus reaching phase
+            if info_container.method == "gdm-score-bands":
+                # Seems like every user wants to move on.
+                statement = select(GroupIteration).where(GroupIteration.group_id == group.id)
+                iterations: list[GroupIteration] = session.exec(statement).all()
+                state: list[SCOREBandsGDMResult] = [
+                    iteration.info_container.score_bands_result
+                    for iteration in iterations
+                    if iteration.info_container.method == "gdm-score-bands"
+                ]
+
+                # USE Bhupinder's score bands stuff.
+                score_bands_config = SCOREBandsGDMConfig(
+                    score_bands_config=info_container.score_bands_config.score_bands_config,
+                    from_iteration=state[-1].iteration,  # the ID from the latest iteration.
+                )
+
+                # Get them discrete reprs
+                discrete_repr = self.discrete_representation
+
+                # Make sure that there are enough solutions for re-clustering
+                votes = group_iteration.info_container.user_votes
+                winners = consensus_rule(votes, score_bands_config.minimum_votes)
+                relevant_ids = state[-1].relevant_ids
+                clustering = state[-1].score_bands_result.clusters
+
+                # threshold to decide whether to move to the decision phase
+                solution_number_threshold = 10
+
+                if (
+                    len([x[0] for x in zip(relevant_ids, clustering, strict=True) if x[1] in winners])
+                    <= solution_number_threshold
+                ):
+                    # There are less than 10 solutions, so move to the decision phase.
+
+                    obj_keys = list(discrete_repr.objective_values)
+                    var_keys = list(discrete_repr.variable_values)
+
+                    objs = pl.DataFrame(discrete_repr.objective_values).with_row_index(name="index_")
+                    varis = pl.DataFrame(discrete_repr.variable_values).with_row_index(name="index_")
+                    indices = pl.DataFrame({"index_": relevant_ids, "cluster_": clustering}).filter(
+                        pl.col("cluster_").is_in(winners)
+                    )
+                    objs = indices.join(other=objs, how="left", left_on="index_", right_on="index_").select(obj_keys)
+                    varis = indices.join(other=varis, how="left", left_on="index_", right_on="index_").select(var_keys)
+
+                    info_container = GDMSCOREBandFinalSelection(
+                        user_votes={},
+                        user_confirms=[],
+                        solution_variables=varis.to_dict(),
+                        solution_objectives=objs.to_dict(),
+                        winner_solution_variables={},
+                        winner_solution_objectives={},
+                    )
+
+                else:
+                    # Case in which we have more than 10 solutions
+                    discrete_repr = discrete_repr.objective_values
+                    objective_keys = list(discrete_repr)
+                    objs = pl.DataFrame(discrete_repr).with_row_index()
+                    objs = objs.select(objective_keys)
+
+                    result: list[SCOREBandsGDMResult] = score_bands_gdm(
+                        data=objs, config=score_bands_config, state=state, votes=votes
+                    )
+
+                    # store necessary data to the database. Currently all
+                    # "voting" related is null bc no voting has happened
+                    info_container = GDMSCOREBandInformation(
+                        user_votes={},
+                        user_confirms=[],
+                        score_bands_config=score_bands_config,
+                        score_bands_result=result[-1],
+                    )
+
+                # Add group iteration and related stuff, then set new iteration to head.
+                new_iteration: GroupIteration = GroupIteration(
+                    group_id=group.id,
+                    problem_id=group.problem_id,
+                    info_container=info_container,
+                    notified={},
+                    state_id=None,
+                    parent_id=group.head_iteration_id,
+                    parent=group_iteration,
+                )
+
+                session.add(new_iteration)
+                session.commit()
+                session.refresh(new_iteration)
+
+                group.head_iteration_id = new_iteration.id
+                session.add(group)
+                session.commit()
+                session.refresh(group)
+
+                await self.broadcast("UPDATE: A new iteration has begun.")
+
+            # We're in decision phase.
+            elif info_container.method == "gdm-score-bands-final":
+                winner = majority_rule(info_container.user_votes)  # A different rule perhaps?
+                # if len(winners) > 1:
+                #    pass # TODO: minimize distance or whatever
+
+                varis = info_container.solution_variables
+                vari_keys = list(varis)
+                objs = info_container.solution_objectives
+                obj_keys = list(objs)
+
+                vari_d = {}
+                for key in vari_keys:
+                    vari_d[key] = varis[key][winner]
+
+                obj_d = {}
+                for key in obj_keys:
+                    obj_d[key] = objs[key][winner]
+
+                info_container.winner_solution_variables = vari_d
+                info_container.winner_solution_objectives = obj_d
+
+                group_iteration.info_container = info_container
+
+                session.add(group_iteration)
+                session.commit()
+
+    async def revert(self, user: User, group: Group, session: Session, group_iteration_number: int):
+        """Revert to a different iteration.
+
+        Args:
+            user (User): Current user
+            group (Group): Current group
+            session (Session): database session
+            group_iteration_number (int): the group iteration to which we want to revert.
+
+        Raises:
+            ManagerError
+        """
+        async with self.lock:
+            group_iteration = session.exec(
+                select(GroupIteration).where(GroupIteration.id == group.head_iteration_id)
+            ).first()
+            if group_iteration is None:
+                raise ManagerError("No group iterations! Did you initialize this group?")
+
+            statement = select(GroupIteration).where(GroupIteration.group_id == group.id)
+            iterations: list[GroupIteration] = session.exec(statement).all()
+
+            target_group_iteration: GroupIteration = next([i for i in iterations if i.id == group_iteration_number])
+            if target_group_iteration.info_container.method == "gdm-score-bands-final":
+                raise ManagerError("We can only revert to a score bands iteration.")
+
+            state: list[SCOREBandsGDMResult] = [
+                iteration.info_container.score_bands_result
+                for iteration in iterations
+                if iteration.info_container.method == "gdm-score-bands"
+            ]
+
+            # The parent ID will be the latest state's iteration
+            prev_id = state[-1].iteration
+            # The results remain the same as in the target
+            result = target_group_iteration.info_container.score_bands_result
+            result.previous_iteration = prev_id
+            result.iteration = prev_id + 1
+
+            # Essentially create a new iteration.
+            info_container = GDMSCOREBandInformation(
+                user_votes={},
+                user_confirms=[],
+                score_bands_config=target_group_iteration.info_container.score_bands_config,
+                score_bands_result=result,
+            )
+
+            # Add group iteration and related stuff, then set new iteration to head.
+            new_iteration: GroupIteration = GroupIteration(
+                group_id=group.id,
+                problem_id=group.problem_id,
+                info_container=info_container,
+                notified={},
+                state_id=None,
+                parent_id=group.head_iteration_id,
+                parent=group_iteration,
+            )
+
+            session.add(new_iteration)
+            session.commit()
+            session.refresh(new_iteration)
+
+            group.head_iteration_id = new_iteration.id
+            session.add(group)
+            session.commit()
+            session.refresh(group)
+
+            await self.broadcast("UPDATE: Iteration reverted.")
+
+    async def configure(self, group: Group, config: SCOREBandsGDMConfig, session: Session):
+        """Configure the SCORE Bands process.
+
+        Args:
+            user (User): Actually might not be necessary...
+            group (Group): The group whom the configuration concerns.
+            config (SCOREBandsGDMConfig): The configuration.
+            session (Session): The database session.
+        """
+        async with self.lock:
+            group_iteration = session.exec(
+                select(GroupIteration).where(GroupIteration.id == group.head_iteration_id)
+            ).first()
+            if group_iteration is None:
+                raise ManagerError("No group iterations! Did you initialize this group?")
+
+            if group_iteration.info_container.method == "gdm-score-bands-final":
+                raise ManagerError("Cannot reconfigure in a non SCORE Bands phase!")
+
+            statement = select(GroupIteration).where(GroupIteration.group_id == group.id)
+            iterations: list[GroupIteration] = session.exec(statement).all()
+            state: list[SCOREBandsGDMResult] = [
+                iteration.info_container.score_bands_result
+                for iteration in iterations
+                if iteration.info_container.method == "gdm-score-bands"
+            ]
+
+            relevant_indices = state[-1].relevant_ids
+            iteration_number = state[-1].iteration
+
+            index_df = pl.DataFrame({"index": relevant_indices})
+
+            discrete_repr = self.discrete_representation.objective_values
+            objective_keys = list(discrete_repr)
+            objs_df = pl.DataFrame(discrete_repr).with_row_index()
+
+            # Filter
+            objs_df = index_df.join(how="left", left_on="index", right_on="index", other=objs_df)
+
+            objs_df = objs_df.select(objective_keys)
+
+            score_bands_result = score_json(data=objs_df, options=config.score_bands_config)
+            score_bands_gdm_result = SCOREBandsGDMResult(
+                score_bands_result=score_bands_result,
+                relevant_ids=relevant_indices,
+                iteration=iteration_number + 1,
+                previous_iteration=iteration_number,
+            )
+
+            # Essentially create a new iteration.
+            info_container = GDMSCOREBandInformation(
+                user_votes={}, user_confirms=[], score_bands_config=config, score_bands_result=score_bands_gdm_result
+            )
+
+            # Add group iteration and related stuff, then set new iteration to head.
+            new_iteration: GroupIteration = GroupIteration(
+                group_id=group.id,
+                problem_id=group.problem_id,
+                info_container=info_container,
+                notified={},
+                state_id=None,
+                parent_id=group.head_iteration_id,
+                parent=group_iteration,
+            )
+
+            session.add(new_iteration)
+            session.commit()
+            session.refresh(new_iteration)
+
+            group.head_iteration_id = new_iteration.id
+            session.add(group)
+            session.commit()
+            session.refresh(group)
+
+            await self.broadcast("UPDATE: Reconfigured SCORE Bands.")