Merge pull request #7 from outerl/deterministic-alt-sample-randoms

Deterministic alt sample randoms

Author: janzill

.github/workflows/core_tests.yml

@@ -141,6 +141,7 @@ jobs:
       matrix:
         region:
           - prototype_mtc
+          - prototype_arc
           - placeholder_psrc
           - prototype_marin
           - prototype_mtc_extended

activitysim/abm/models/location_choice.py

@@ -18,6 +18,7 @@
 from activitysim.core.exceptions import DuplicateWorkflowTableError
 from activitysim.core.interaction_sample import interaction_sample
 from activitysim.core.interaction_sample_simulate import interaction_sample_simulate
+from activitysim.core.logit import AltsContext
 from activitysim.core.util import reindex
 
 """
@@ -603,6 +604,7 @@ def run_location_simulate(
     chunk_tag,
     trace_label,
     skip_choice=False,
+    alts_context: AltsContext | None = None,
 ):
     """
     run location model on location_sample annotated with mode_choice logsum
@@ -712,6 +714,7 @@ def run_location_simulate(
         compute_settings=model_settings.compute_settings.subcomponent_settings(
             "simulate"
         ),
+        alts_context=alts_context,
     )
 
     if not want_logsums:
@@ -737,6 +740,7 @@ def run_location_choice(
     chunk_tag,
     trace_label,
     skip_choice=False,
+    alts_context: AltsContext | None = None,
 ):
     """
     Run the three-part location choice algorithm to generate a location choice for each chooser
@@ -756,6 +760,8 @@ def run_location_choice(
     model_settings : dict
     chunk_size : int
     trace_label : str
+    skip_choice : bool
+    alts_context : AltsContext or None
 
     Returns
     -------
@@ -788,6 +794,9 @@ def run_location_choice(
         if choosers.shape[0] == 0:
             logger.info(f"{trace_label} skipping segment {segment_name}: no choosers")
             continue
+        # dest_size_terms contains 0-attraction zones so using this directly here, important for stable error terms
+        # when a zone goes from 0 base -> nonzero project
+        alts_context = AltsContext.from_series(dest_size_terms.index)
 
         # - location_sample
         location_sample_df = run_location_sample(
@@ -841,6 +850,7 @@ def run_location_choice(
                 trace_label, "simulate.%s" % segment_name
             ),
             skip_choice=skip_choice,
+            alts_context=alts_context,
         )
 
         if estimator:

activitysim/abm/models/parking_location_choice.py

@@ -21,6 +21,7 @@
 from activitysim.core.configuration.base import PreprocessorSettings
 from activitysim.core.configuration.logit import LogitComponentSettings
 from activitysim.core.interaction_sample_simulate import interaction_sample_simulate
+from activitysim.core.logit import AltsContext
 from activitysim.core.tracing import print_elapsed_time
 from activitysim.core.util import assign_in_place, drop_unused_columns
 from activitysim.core.exceptions import DuplicateWorkflowTableError
@@ -112,6 +113,7 @@ def parking_destination_simulate(
     chunk_size,
     trace_hh_id,
     trace_label,
+    alts_context: AltsContext | None = None,
 ):
     """
     Chose destination from destination_sample (with od_logsum and dp_logsum columns added)
@@ -150,6 +152,7 @@ def parking_destination_simulate(
         trace_label=trace_label,
         trace_choice_name="parking_loc",
         explicit_chunk_size=model_settings.explicit_chunk,
+        alts_context=alts_context,
     )
 
     # drop any failed zero_prob destinations
@@ -211,6 +214,9 @@ def choose_parking_location(
     )
     destination_sample.index = np.repeat(trips.index.values, len(alternatives))
     destination_sample.index.name = trips.index.name
+    # use full land_use index to ensure AltsContext spans full range of potential zones
+    land_use = state.get_dataframe("land_use")
+    alts_context = AltsContext.from_series(land_use.index)
 
     destinations = parking_destination_simulate(
         state,
@@ -223,6 +229,7 @@ def choose_parking_location(
         chunk_size=chunk_size,
         trace_hh_id=trace_hh_id,
         trace_label=trace_label,
+        alts_context=alts_context,
     )
 
     if want_sample_table:

activitysim/abm/models/trip_destination.py

@@ -32,6 +32,7 @@
 from activitysim.core.configuration.logit import LocationComponentSettings
 from activitysim.core.interaction_sample import interaction_sample
 from activitysim.core.interaction_sample_simulate import interaction_sample_simulate
+from activitysim.core.logit import AltsContext
 from activitysim.core.skim_dictionary import DataFrameMatrix
 from activitysim.core.tracing import print_elapsed_time
 from activitysim.core.util import assign_in_place, reindex
@@ -950,6 +951,7 @@ def trip_destination_simulate(
     skim_hotel,
     estimator,
     trace_label,
+    alts_context: AltsContext | None = None,
 ):
     """
     Chose destination from destination_sample (with od_logsum and dp_logsum columns added)
@@ -1036,6 +1038,7 @@ def trip_destination_simulate(
         trace_choice_name="trip_dest",
         estimator=estimator,
         explicit_chunk_size=model_settings.explicit_chunk,
+        alts_context=alts_context,
     )
 
     if not want_logsums:
@@ -1080,6 +1083,10 @@ def choose_trip_destination(
 
     t0 = print_elapsed_time()
 
+    # use full index (including zero-size zones) to ensure stable random results
+    # fetch alts_context early so we don't worry about mutating alternatives first
+    alts_context = AltsContext.from_series(alternatives.index)
+
     # - trip_destination_sample
     destination_sample = trip_destination_sample(
         state,
@@ -1126,7 +1133,6 @@ def choose_trip_destination(
         destination_sample["dp_logsum"] = 0.0
 
     t0 = print_elapsed_time("%s.compute_logsums" % trace_label, t0, debug=True)
-
     destinations = trip_destination_simulate(
         state,
         primary_purpose=primary_purpose,
@@ -1138,6 +1144,7 @@ def choose_trip_destination(
         skim_hotel=skim_hotel,
         estimator=estimator,
         trace_label=trace_label,
+        alts_context=alts_context,
     )
 
     dropped_trips = ~trips.index.isin(destinations.index)

activitysim/abm/models/trip_scheduling_choice.py

@@ -279,6 +279,15 @@ def run_trip_scheduling_choice(
         ) in chunk.adaptive_chunked_choosers(state, indirect_tours, trace_label):
             # Sort the choosers and get the schedule alternatives
             choosers = choosers.sort_index()
+            # FIXME-EET: For explicit error term choices, we need a stable alternative ID. Currently, we use
+            # SCHEDULE_ID, which justs enumerates all schedule alternatives, of which there are choosers times
+            # alternative, in the order they are processed, which depends on if there stops on outward/return leg.
+            # We might want to change SCHEDULE_ID to a fixed pattern of all possible combinations of
+            # (outbound, main, inbound) duration for the maximum possible tour duration (max time window). For
+            # 30min intervals, this leads to 1225 alternatives and therefore reasonable memory-wise for random numbers.
+            # It looks like all that would need to change for this is the generation of the schedule alternatives and
+            # the lookup of choices as elements in schedule after simulation because choosers are indexed by tour_id.
+
             schedules = generate_schedule_alternatives(choosers).sort_index()
 
             # preprocessing alternatives

activitysim/abm/models/util/tour_destination.py

@@ -21,6 +21,7 @@
 from activitysim.core.configuration.logit import TourLocationComponentSettings
 from activitysim.core.interaction_sample import interaction_sample
 from activitysim.core.interaction_sample_simulate import interaction_sample_simulate
+from activitysim.core.logit import AltsContext
 from activitysim.core.util import reindex
 
 logger = logging.getLogger(__name__)
@@ -873,6 +874,10 @@ def run_destination_simulate(
     state.tracing.dump_df(DUMP, choosers, trace_label, "choosers")
 
     log_alt_losers = state.settings.log_alt_losers
+    # use full land_use index to ensure AltsContext spans full range of potential destinations
+    # (maintains stable random number generation even if zones flip zero/non-zero size)
+    land_use = state.get_dataframe("land_use")
+    alts_context = AltsContext.from_series(land_use.index)
 
     choices = interaction_sample_simulate(
         state,
@@ -891,6 +896,7 @@ def run_destination_simulate(
         estimator=estimator,
         skip_choice=skip_choice,
         compute_settings=model_settings.compute_settings,
+        alts_context=alts_context,
     )
 
     if not want_logsums:

activitysim/abm/models/util/vectorize_tour_scheduling.py

@@ -17,6 +17,7 @@
 from activitysim.core.configuration.base import ComputeSettings, PreprocessorSettings
 from activitysim.core.configuration.logit import LogitComponentSettings
 from activitysim.core.interaction_sample_simulate import interaction_sample_simulate
+from activitysim.core.logit import AltsContext
 from activitysim.core.util import reindex
 
 logger = logging.getLogger(__name__)
@@ -849,6 +850,9 @@ def _schedule_tours(
         estimator.write_interaction_sample_alternatives(alt_tdd)
 
     log_alt_losers = state.settings.log_alt_losers
+    # use full TDD alternatives index to ensure AltsContext spans full range of potential slots
+    tdd_alts = state.get_injectable("tdd_alts")
+    alts_context = AltsContext.from_series(tdd_alts.index)
 
     choices = interaction_sample_simulate(
         state,
@@ -862,6 +866,7 @@ def _schedule_tours(
         trace_label=tour_trace_label,
         estimator=estimator,
         compute_settings=compute_settings,
+        alts_context=alts_context,
     )
     chunk_sizer.log_df(tour_trace_label, "choices", choices)
 

activitysim/abm/test/test_misc/test_trip_scheduling_choice.py

@@ -1,9 +1,11 @@
-import numpy as np
-import pandas as pd
-import pytest
+from __future__ import annotations
+
 import os
 from pathlib import Path
 
+import numpy as np
+import pandas as pd
+import pytest
 
 from activitysim.abm.models import trip_scheduling_choice as tsc
 from activitysim.abm.tables.skims import skim_dict

activitysim/core/interaction_sample_simulate.py

@@ -9,8 +9,9 @@
 
 from activitysim.core import chunk, interaction_simulate, logit, tracing, util, workflow
 from activitysim.core.configuration.base import ComputeSettings
-from activitysim.core.simulate import set_skim_wrapper_targets
 from activitysim.core.exceptions import SegmentedSpecificationError
+from activitysim.core.logit import AltsContext
+from activitysim.core.simulate import set_skim_wrapper_targets
 
 logger = logging.getLogger(__name__)
 
@@ -34,6 +35,7 @@ def _interaction_sample_simulate(
     *,
     chunk_sizer: chunk.ChunkSizer,
     compute_settings: ComputeSettings | None = None,
+    alts_context: AltsContext | None = None,
 ):
     """
     Run a MNL simulation in the situation in which alternatives must
@@ -220,9 +222,6 @@ def _interaction_sample_simulate(
     )
     chunk_sizer.log_df(trace_label, "interaction_utilities", interaction_utilities)
 
-    del interaction_df
-    chunk_sizer.log_df(trace_label, "interaction_df", None)
-
     if have_trace_targets:
         state.tracing.trace_interaction_eval_results(
             trace_eval_results,
@@ -264,19 +263,29 @@ def _interaction_sample_simulate(
 
     # insert the zero-prob utilities to pad each alternative set to same size
     padded_utilities = np.insert(interaction_utilities.utility.values, inserts, -999)
+    padded_alt_nrs = np.insert(interaction_df[choice_column], inserts, -999)
     chunk_sizer.log_df(trace_label, "padded_utilities", padded_utilities)
-    del inserts
 
-    del interaction_utilities
-    chunk_sizer.log_df(trace_label, "interaction_utilities", None)
+    del interaction_df
+    chunk_sizer.log_df(trace_label, "interaction_df", None)
+
+    del inserts
 
     # reshape to array with one row per chooser, one column per alternative
     padded_utilities = padded_utilities.reshape(-1, max_sample_count)
+    padded_alt_nrs = padded_alt_nrs.reshape(-1, max_sample_count)
 
     # convert to a dataframe with one row per chooser and one column per alternative
     utilities_df = pd.DataFrame(padded_utilities, index=choosers.index)
     chunk_sizer.log_df(trace_label, "utilities_df", utilities_df)
 
+    # alt_nrs_df has columns for each alt in the choice set, with values indicating which alt_id
+    # they correspond to (as opposed to the 0-n index implied by the column number).
+    if alts_context is not None:
+        alt_nrs_df = pd.DataFrame(padded_alt_nrs, index=choosers.index)
+    else:
+        alt_nrs_df = None  # if we don't provide the number of dense alternatives, assume that we'll use the old approach
+
     del padded_utilities
     chunk_sizer.log_df(trace_label, "padded_utilities", None)
 
@@ -320,7 +329,12 @@ def _interaction_sample_simulate(
         # positions is series with the chosen alternative represented as a column index in utilities_df
         # which is an integer between zero and num alternatives in the alternative sample
         positions, rands = logit.make_choices_utility_based(
-            state, utilities_df, trace_label=trace_label, trace_choosers=choosers
+            state,
+            utilities_df,
+            trace_label=trace_label,
+            trace_choosers=choosers,
+            alts_context=alts_context,
+            alt_nrs_df=alt_nrs_df,
         )
 
         del utilities_df
@@ -451,6 +465,7 @@ def interaction_sample_simulate(
     skip_choice=False,
     explicit_chunk_size=0,
     *,
+    alts_context: AltsContext | None = None,
     compute_settings: ComputeSettings | None = None,
 ):
     """
@@ -496,6 +511,12 @@ def interaction_sample_simulate(
     explicit_chunk_size : float, optional
         If > 0, specifies the chunk size to use when chunking the interaction
         simulation. If < 1, specifies the fraction of the total number of choosers.
+    alts_context: AltsContext, optional
+        Representation of the full alternatives domain (min and max alternative id)
+        in the absence of sampling.
+        This is used with EET simulation to ensure consistent random numbers across the whole alternative set
+        ( as the sampled set may change between base and project). When not provided,
+        EET with integer-coded choice ids will raise an error.
 
     Returns
     -------
@@ -517,6 +538,18 @@ def interaction_sample_simulate(
     trace_label = tracing.extend_trace_label(trace_label, "interaction_sample_simulate")
     chunk_tag = chunk_tag or trace_label
 
+    if state.settings.use_explicit_error_terms:
+        choice_ids_are_int = pd.api.types.is_integer_dtype(alternatives[choice_column])
+        if alts_context is None and choice_ids_are_int:
+            raise ValueError(
+                "alts_context is required for interaction_sample_simulate when "
+                "use_explicit_error_terms is True and choice_column is integer-coded"
+            )
+        if alts_context is not None and not choice_ids_are_int:
+            raise ValueError(
+                "alts_context can only be used with integer-coded choice_column values"
+            )
+
     result_list = []
     for (
         i,
@@ -551,6 +584,7 @@ def interaction_sample_simulate(
             skip_choice,
             chunk_sizer=chunk_sizer,
             compute_settings=compute_settings,
+            alts_context=alts_context,
         )
 
         result_list.append(choices)