add stop frequency

.gitignore

@@ -13,3 +13,4 @@ utilities/sandag-activitysim/configs/*
 utilities/sandag-activitysim/output/*
 utilities/sandag-activitysim/scripts/.Rproj.user/*
 utilities/sandag-activitysim/scripts/.Rhistory
+.Rproj.user

utilities/sandag-activitysim/scripts/stop-frequency.Rmd

@@ -0,0 +1,154 @@
+---
+title: "Stop Frequency"
+output: html_notebook
+---
+
+# Overhead
+```{r overhead, include = FALSE}
+packages_vector <- c("tidyverse",
+                     "kableExtra")
+
+need_to_install <- packages_vector[!(packages_vector %in% installed.packages()[,"Package"])]
+
+if (length(need_to_install)) install.packages(need_to_install)
+
+for (package in packages_vector) {
+  library(package, character.only = TRUE)
+}
+
+```
+
+# Remote I/O
+```{r remote-io}
+interim_dir <- "../output/"
+
+tour_filename <- paste0(interim_dir, "final_tours.csv")
+
+```
+
+# Parameters
+```{r parameters}
+
+```
+
+
+# Data Reads
+```{r read}
+tour_df <- read_csv(tour_filename, col_types = cols(
+  tour_id = col_double(),
+  person_id = col_double(),
+  tour_type = col_character(),
+  tour_type_count = col_double(),
+  tour_type_num = col_double(),
+  tour_num = col_double(),
+  tour_count = col_double(),
+  tour_category = col_character(),
+  number_of_participants = col_double(),
+  destination = col_double(),
+  origin = col_double(),
+  household_id = col_double(),
+  start = col_double(),
+  end = col_double(),
+  duration = col_double(),
+  school_esc_outbound = col_character(),
+  school_esc_inbound = col_character(),
+  num_escortees = col_double(),
+  tdd = col_double(),
+  tour_id_temp = col_double(),
+  composition = col_character(),
+  is_external_tour = col_logical(),
+  is_internal_tour = col_logical(),
+  destination_logsum = col_double(),
+  vehicle_occup_1 = col_character(),
+  vehicle_occup_2 = col_character(),
+  vehicle_occup_3.5 = col_character(),
+  tour_mode = col_character(),
+  mode_choice_logsum = col_double(),
+  selected_vehicle = col_character(),
+  atwork_subtour_frequency = col_character(),
+  parent_tour_id = col_double(),
+  stop_frequency = col_character(),
+  primary_purpose = col_character(),
+  model = col_character()
+))
+```
+
+# Reductions 
+```{r reductions}
+working_df <- tour_df %>%
+  select(tour_id, person_id, tour_category, tour_type, stop_frequency) %>%
+  group_by(tour_category, tour_type, stop_frequency) %>%
+  summarise(count = n(), .groups = "drop") %>%
+  group_by(tour_category, tour_type) %>%
+  mutate(share = count/sum(count)) %>%
+  ungroup()
+```
+
+Examine four selected purposes to see if examples are logical. 
+
+1. Work tours
+```{r}
+summary_df <- working_df %>%
+  filter(tour_category == "mandatory" & tour_type == "work") %>%
+  arrange(-share)
+
+summary_df %>%
+  kbl() %>%
+  kable_styling()
+```
+
+This makes sense. The dominate outcome is 0 out, 0 in, followed by 0 out, 1 in. It is somewhat illogical that 0 out 3 in is more likely than 0 out, 2 in. Check the UEC for this. 
+
+1. At-work eating tours
+```{r}
+summary_df <- working_df %>%
+  filter(tour_category == "atwork" & tour_type == "eat") %>%
+  arrange(-share)
+
+summary_df %>%
+  kbl() %>%
+  kable_styling()
+```
+
+A dominate share of 0 out, 0 in, as expected.
+
+3. Shopping tours
+```{r}
+summary_df <- working_df %>%
+  filter(tour_category == "non_mandatory" & tour_type == "shopping") %>%
+  arrange(-share)
+
+summary_df %>%
+  kbl() %>%
+  kable_styling()
+```
+
+0 out, 0 in is a bit higher than expected, but the in/out pairs (1 out, 0 in and 0 out, 1 in) are roughly balanced, as expected. 
+
+4. Joint social
+```{r}
+summary_df <- working_df %>%
+  filter(tour_category == "joint" & tour_type == "social") %>%
+  arrange(-share)
+
+summary_df %>%
+  kbl() %>%
+  kable_styling()
+```
+
+These outcomes appear to be a bit overfit to the data, with 3 out, 0 in being the most likely. Perhaps this is being addressed in the on-going calibration effort.  
+
+
+# Findings
+The models appear to be implemented correctly. Though a minor issue, the models could benefit from more thoughtful calibration, though the impact would be relatively minor. For example, the base coefficients for stops on social tours on the outbound trip is `1.323` -- it does not vary across the alternatives. During calibration of SANDAG's ABM2, it appears a constant of `3.226` was applied to the `3 out, 0 in` alternative, but a smaller constant of `1.75` to the `2 out, 0 in` alternative. The current calibration adjusts both of these constants down, but not enough to account for the gap. It's unlikely that tours with 3 outbound stops are actually more likely than those with 2. MTC should consider cleaning up the calibration. 
+
+ActivitySim files reviewed include:
+
+- `stop_frequency_alternatives.csv`
+- `stop_frequency_annotate_tours_preprocessor.csv`
+- `stop_frequency_[purpose].csv`, where purpose is `at_work`, `eatout`, `escort`, `othdiscr`, `othmaint`, `school`, `shopping`, `social`, `univ`, and `work`
+
+- `stop_frequency_coefficients_[purpose]`, where purpose is the same as above.
+- `stop_frequency.yaml`
+
+