categorical time period dtype

activitysim/abm/models/parking_location_choice.py

@@ -318,7 +318,7 @@ def parking_location(
         if "trip_period" not in trips_merged_df:
             # TODO: resolve this to the skim time period index not the label, it will be faster
             trips_merged_df["trip_period"] = network_los.skim_time_period_label(
-                trips_merged_df[proposed_trip_departure_period]
+                trips_merged_df[proposed_trip_departure_period], as_cat=True
             )
         model_settings["TRIP_DEPARTURE_PERIOD"] = "trip_period"
 

activitysim/abm/models/trip_mode_choice.py

@@ -73,7 +73,7 @@ def trip_mode_choice(
     # setup skim keys
     assert "trip_period" not in trips_merged
     trips_merged["trip_period"] = network_los.skim_time_period_label(
-        trips_merged.depart
+        trips_merged.depart, as_cat=True
     )
 
     orig_col = "origin"

activitysim/abm/models/util/logsums.py

@@ -75,10 +75,10 @@ def compute_logsums(
     # FIXME - are we ok with altering choosers (so caller doesn't have to set these)?
     if (in_period_col is not None) and (out_period_col is not None):
         choosers["in_period"] = network_los.skim_time_period_label(
-            choosers[in_period_col]
+            choosers[in_period_col], as_cat=True
         )
         choosers["out_period"] = network_los.skim_time_period_label(
-            choosers[out_period_col]
+            choosers[out_period_col], as_cat=True
         )
     elif ("in_period" not in choosers.columns) and (
         "out_period" not in choosers.columns
@@ -92,17 +92,21 @@ def compute_logsums(
                 and tour_purpose in model_settings["OUT_PERIOD"]
             ):
                 choosers["in_period"] = network_los.skim_time_period_label(
-                    model_settings["IN_PERIOD"][tour_purpose]
+                    model_settings["IN_PERIOD"][tour_purpose],
+                    as_cat=True,
+                    broadcast_to=choosers.index,
                 )
                 choosers["out_period"] = network_los.skim_time_period_label(
-                    model_settings["OUT_PERIOD"][tour_purpose]
+                    model_settings["OUT_PERIOD"][tour_purpose],
+                    as_cat=True,
+                    broadcast_to=choosers.index,
                 )
         else:
             choosers["in_period"] = network_los.skim_time_period_label(
-                model_settings["IN_PERIOD"]
+                model_settings["IN_PERIOD"], as_cat=True, broadcast_to=choosers.index
             )
             choosers["out_period"] = network_los.skim_time_period_label(
-                model_settings["OUT_PERIOD"]
+                model_settings["OUT_PERIOD"], as_cat=True, broadcast_to=choosers.index
             )
     else:
         logger.error("Choosers table already has columns 'in_period' and 'out_period'.")

activitysim/abm/models/util/mode.py

@@ -131,8 +131,12 @@ def run_tour_mode_choice_simulate(
     assert ("in_period" not in choosers) and ("out_period" not in choosers)
     in_time = skims["in_time_col_name"]
     out_time = skims["out_time_col_name"]
-    choosers["in_period"] = network_los.skim_time_period_label(choosers[in_time])
-    choosers["out_period"] = network_los.skim_time_period_label(choosers[out_time])
+    choosers["in_period"] = network_los.skim_time_period_label(
+        choosers[in_time], as_cat=True
+    )
+    choosers["out_period"] = network_los.skim_time_period_label(
+        choosers[out_time], as_cat=True
+    )
 
     expressions.annotate_preprocessors(
         state, choosers, locals_dict, skims, model_settings, trace_label

activitysim/abm/models/util/vectorize_tour_scheduling.py

@@ -185,6 +185,12 @@ def dedupe_alt_tdd(state: workflow.State, alt_tdd, tour_purpose, trace_label):
 
     logger.info("tdd_alt_segments specified for representative logsums")
 
+    if tdd_segments is not None:
+        # apply categorical dtypes
+        tdd_segments["time_period"] = tdd_segments["time_period"].astype(
+            alt_tdd["out_period"].dtype
+        )
+
     with chunk.chunk_log(
         state, tracing.extend_trace_label(trace_label, "dedupe_alt_tdd")
     ) as chunk_sizer:
@@ -328,11 +334,12 @@ def compute_tour_scheduling_logsums(
     assert "out_period" not in alt_tdd
     assert "in_period" not in alt_tdd
 
-    # FIXME:MEMORY
-    #  These two lines each generate a massive array of strings,
-    #  using a bunch of RAM and slowing things down.
-    alt_tdd["out_period"] = network_los.skim_time_period_label(alt_tdd["start"])
-    alt_tdd["in_period"] = network_los.skim_time_period_label(alt_tdd["end"])
+    alt_tdd["out_period"] = network_los.skim_time_period_label(
+        alt_tdd["start"], as_cat=True
+    )
+    alt_tdd["in_period"] = network_los.skim_time_period_label(
+        alt_tdd["end"], as_cat=True
+    )
 
     alt_tdd["duration"] = alt_tdd["end"] - alt_tdd["start"]
 
@@ -383,17 +390,28 @@ def compute_tour_scheduling_logsums(
 
         # tracing.log_runtime(model_name=trace_label, start_time=t0)
 
-        # redupe - join the alt_tdd_period logsums to alt_tdd to get logsums for alt_tdd
-        logsums = (
-            pd.merge(
-                alt_tdd.reset_index(),
-                deduped_alt_tdds.reset_index(),
-                on=[index_name] + redupe_columns,
-                how="left",
-            )
-            .set_index(index_name)
-            .logsums
-        )
+        logsums = pd.Series(data=0, index=alt_tdd.index, dtype=np.float64)
+        left_on = [alt_tdd.index]
+        right_on = [deduped_alt_tdds.index]
+        for i in redupe_columns:
+            if (
+                alt_tdd[i].dtype == "category"
+                and alt_tdd[i].dtype.ordered
+                and alt_tdd[i].dtype == deduped_alt_tdds[i].dtype
+            ):
+                left_on += [alt_tdd[i].cat.codes]
+                right_on += [deduped_alt_tdds[i].cat.codes]
+            else:
+                left_on += [alt_tdd[i].to_numpy()]
+                right_on += [deduped_alt_tdds[i].to_numpy()]
+
+        logsums.iloc[:] = pd.merge(
+            pd.DataFrame(index=alt_tdd.index),
+            deduped_alt_tdds.logsums,
+            left_on=left_on,
+            right_on=right_on,
+            how="left",
+        ).logsums.to_numpy()
         chunk_sizer.log_df(trace_label, "logsums", logsums)
 
         del deduped_alt_tdds

activitysim/core/los.py

@@ -845,7 +845,9 @@ def get_tappairs3d(self, otap, dtap, dim3, key):
 
         return s.values
 
-    def skim_time_period_label(self, time_period, fillna=None):
+    def skim_time_period_label(
+        self, time_period, fillna=None, as_cat=False, broadcast_to=None
+    ):
         """
         convert time period times to skim time period labels (e.g. 9 -> 'AM')
 
@@ -888,6 +890,12 @@ def skim_time_period_label(self, time_period, fillna=None):
                 result = self.skim_time_periods["labels"].get(bin, default=default)
             else:
                 result = self.skim_time_periods["labels"][bin]
+            if broadcast_to is not None and as_cat:
+                result = pd.Series(
+                    data=result,
+                    index=broadcast_to,
+                    dtype=self.skim_dicts["taz"].time_label_dtype,
+                )
         else:
             result = pd.cut(
                 time_period,
@@ -898,8 +906,10 @@ def skim_time_period_label(self, time_period, fillna=None):
             if fillna is not None:
                 default = self.skim_time_periods["labels"][fillna]
                 result = result.fillna(default)
-            result = result.astype(str)
-
+            if as_cat:
+                result = result.astype(self.skim_dicts["taz"].time_label_dtype)
+            else:
+                result = result.astype(str)
         return result
 
     def get_tazs(self, state):

activitysim/core/skim_dataset.py

@@ -33,6 +33,10 @@ def __init__(self, dataset):
         self.time_map = {
             j: i for i, j in enumerate(self.dataset.indexes["time_period"])
         }
+        self.time_label_dtype = pd.api.types.CategoricalDtype(
+            self.dataset.indexes["time_period"],
+            ordered=True,
+        )
         self.usage = set()  # track keys of skims looked up
 
     @property
@@ -184,6 +188,10 @@ def __init__(self, dataset, orig_key, dest_key, time_key=None, *, time_map=None)
             }
         else:
             self.time_map = time_map
+        self.time_label_dtype = pd.api.types.CategoricalDtype(
+            self.dataset.indexes["time_period"],
+            ordered=True,
+        )
 
     @property
     def odim(self):
@@ -246,6 +254,11 @@ def set_df(self, df):
             ):
                 logger.info(f"natural use for time_period={self.time_key}")
                 positions["time_period"] = df[self.time_key]
+            elif (
+                df[self.time_key].dtype == "category"
+                and df[self.time_key].dtype == self.time_label_dtype
+            ):
+                positions["time_period"] = df[self.time_key].cat.codes
             else:
                 logger.info(f"vectorize lookup for time_period={self.time_key}")
                 positions["time_period"] = pd.Series(
@@ -257,11 +270,18 @@ def set_df(self, df):
             self.positions = {}
             for k, v in positions.items():
                 try:
-                    self.positions[k] = v.astype(int)
-                except TypeError:
-                    # possibly some missing values that are not relevant,
-                    # fill with zeros to continue.
-                    self.positions[k] = v.fillna(0).astype(int)
+                    is_int = np.issubdtype(v.dtype, np.integer)
+                except Exception:
+                    is_int = False
+                if is_int:
+                    self.positions[k] = v
+                else:
+                    try:
+                        self.positions[k] = v.astype(int)
+                    except TypeError:
+                        # possibly some missing values that are not relevant,
+                        # fill with zeros to continue.
+                        self.positions[k] = v.fillna(0).astype(int)
         else:
             self.positions = pd.DataFrame(positions).astype(int)