Optimize Barnett's computation multi-day segments

Snakefile

@@ -205,6 +205,9 @@ for provider in config["PHONE_LOCATIONS"]["PROVIDERS"].keys():
             else:
                 raise ValueError("Error: Add PHONE_LOCATIONS (and as many PHONE_SENSORS as you have) to [PHONE_DATA_YIELD][SENSORS] in config.yaml. This is necessary to compute phone_yielded_timestamps (time when the smartphone was sensing data) which is used to resample fused location data (ALL_RESAMPLED and RESAMPLED_FUSED)")
 
+        if provider == "BARNETT":
+            files_to_compute.extend(expand("data/interim/{pid}/phone_locations_barnett_daily.csv", pid=config["PIDS"]))
+
         files_to_compute.extend(expand("data/raw/{pid}/phone_locations_raw.csv", pid=config["PIDS"]))
         files_to_compute.extend(expand("data/interim/{pid}/phone_locations_processed.csv", pid=config["PIDS"]))
         files_to_compute.extend(expand("data/interim/{pid}/phone_locations_processed_with_datetime.csv", pid=config["PIDS"]))

docs/features/phone-locations.md

@@ -33,7 +33,7 @@ These features are based on the original open-source implementation by [Barnett
 
 
 !!! info "Available time segments and platforms"
-    - Available only for segments that start at 00:00:00 and end at 23:59:59 of the same day (daily segments)
+    - Available only for segments that start at 00:00:00 and end at 23:59:59 of the same or a different day (daily, weekly, weekend, etc.)
     - Available for Android and iOS
 
 !!! info "File Sequence"
@@ -78,7 +78,17 @@ Features description for `[PHONE_LOCATIONS][PROVIDERS][BARNETT]` adapted from [B
 |wkenddayrtn                            |       -    | Same as circdnrtn but computed separately for weekends and weekdays.
 
 !!! note "Assumptions/Observations"
-    **Barnett\'s et al features**
+    **Multi day segment features**
+    Barnett's features are only available on time segments that span entire days (00:00:00 to 23:59:59). Such segments can be one-day long (daily) or multi-day (weekly, for example). Multi-day segment features are computed based on daily features summarized the following way:
+
+    - sum for `hometime`, `disttravelled`, `siglocsvisited`, and `minutes_data_used`
+    - max for `maxdiam`, and `maxhomedist`
+    - mean for `rog`, `avgflightlen`, `stdflightlen`, `avgflightdur`, `stdflightdur`, `probpause`, `siglocentropy`, `circdnrtn`, `wkenddayrtn`, and `minsmissing`
+
+    **Computation speed**
+    The process to extract these features can be slow compared to other sensors and providers due to the required simulation.
+
+    **How are these features computed?**
     These features are based on a Pause-Flight model. A pause is defined as a mobility trace (location pings) within a certain duration and distance (by default, 300 seconds and 60 meters). A flight is any mobility trace between two pauses. Data is resampled and imputed before the features are computed. See [Barnett et al](../../citation#barnett-locations) for more information. In RAPIDS, we only expose one parameter for these features (accuracy limit). You can change other parameters in `src/features/phone_locations/barnett/library/MobilityFeatures.R`.
 
     **Significant Locations**

rules/common.smk

@@ -11,6 +11,11 @@ def get_script_language(script_path):
 
 
 # Features.smk #########################################################################################################
+def get_barnett_daily(wildcards):
+    if wildcards.provider_key.upper() == "BARNETT":
+        return "data/interim/{pid}/phone_locations_barnett_daily.csv"
+    return []
+
 def find_features_files(wildcards):
     feature_files = []
     for provider_key, provider in config[(wildcards.sensor_key).upper()]["PROVIDERS"].items():

rules/features.smk

@@ -379,10 +379,22 @@ rule phone_locations_python_features:
     script:
         "../src/features/entry.py"
 
+rule phone_locations_barnett_daily_features:
+    input:
+        sensor_data = "data/interim/{pid}/phone_locations_processed_with_datetime.csv",
+        time_segments_labels = "data/interim/time_segments/{pid}_time_segments_labels.csv",
+    params:
+        provider = lambda wildcards: config["PHONE_LOCATIONS"]["PROVIDERS"]["BARNETT"],
+    output:
+        "data/interim/{pid}/phone_locations_barnett_daily.csv"
+    script:
+        "../src/features/phone_locations/barnett/daily_features.R"
+
 rule phone_locations_r_features:
     input:
         sensor_data = "data/interim/{pid}/phone_locations_processed_with_datetime.csv",
-        time_segments_labels = "data/interim/time_segments/{pid}_time_segments_labels.csv"
+        time_segments_labels = "data/interim/time_segments/{pid}_time_segments_labels.csv",
+        barnett_daily =  get_barnett_daily
     params:
         provider = lambda wildcards: config["PHONE_LOCATIONS"]["PROVIDERS"][wildcards.provider_key.upper()],
         provider_key = "{provider_key}",

src/features/phone_locations/barnett/daily_features.R

@@ -0,0 +1,68 @@
+source("renv/activate.R")
+library("dplyr", warn.conflicts = F)
+library("stringr")
+library("lubridate")
+library("purrr")
+
+# Load Ian Barnett's code. From https://scholar.harvard.edu/ibarnett/software/gpsmobility
+file.sources = list.files(c("src/features/phone_locations/barnett/library"), pattern="*.R$", full.names=TRUE, ignore.case=TRUE)
+output_apply <- sapply(file.sources,source,.GlobalEnv)
+
+create_empty_file <- function(){
+  return(data.frame(local_date= character(), hometime= numeric(), disttravelled= numeric(), rog= numeric(), maxdiam= numeric(), 
+                    maxhomedist= numeric(), siglocsvisited= numeric(), avgflightlen= numeric(), stdflightlen= numeric(), 
+                    avgflightdur= numeric(), stdflightdur= numeric(), probpause= numeric(), siglocentropy= numeric(), minsmissing= numeric(), 
+                    circdnrtn= numeric(), wkenddayrtn= numeric(), minutes_data_used= numeric()
+  ))
+}
+
+barnett_daily_features <- function(snakemake){
+  location_features <- NULL
+  location <- read.csv(snakemake@input[["sensor_data"]], stringsAsFactors = FALSE)
+  segment_labels <- read.csv(snakemake@input[["time_segments_labels"]], stringsAsFactors = FALSE)
+  accuracy_limit <- snakemake@params[["provider"]][["ACCURACY_LIMIT"]]
+  datetime_start_regex = "[0-9]{4}[\\-|\\/][0-9]{2}[\\-|\\/][0-9]{2} 00:00:00"
+  datetime_end_regex = "[0-9]{4}[\\-|\\/][0-9]{2}[\\-|\\/][0-9]{2} 23:59:59"
+  location <- location %>% 
+    filter(accuracy < accuracy_limit) %>% 
+    mutate(is_daily = str_detect(assigned_segments, paste0(".*#", datetime_start_regex, ",", datetime_end_regex, ".*")))
+    
+  
+  if(nrow(location) == 0 || all(location$is_daily == FALSE) || (max(location$timestamp) - min(location$timestamp) < 86400000)){
+    warning("Barnett's location features cannot be computed for data or time segments that do not span one or more entire days (00:00:00 to 23:59:59). Values below point to the problem:",
+            "\nLocation data rows within accuracy: ", nrow(location %>% filter(accuracy < accuracy_limit)),
+            "\nLocation data rows within a daily time segment: ", nrow(filter(location, is_daily)),
+            "\nLocation data time span in days: ", round((max(location$timestamp) - min(location$timestamp)) / 86400000, 2)
+            )
+    location_features <- create_empty_file()  
+  } else{
+    # Count how many minutes of data we use to get location features. Some minutes have multiple fused rows
+    location_minutes_used <- location %>% 
+      group_by(local_date, local_hour) %>% 
+      summarise(n_minutes = n_distinct(local_minute), .groups = 'drop_last') %>% 
+      group_by(local_date) %>% 
+      summarise(minutes_data_used = sum(n_minutes), .groups = 'drop_last') %>% 
+      select(local_date, minutes_data_used)
+    
+    # Select only the columns that the algorithm needs
+    all_timezones <- table(location %>% pull(local_timezone))
+    location <- location %>% select(timestamp, latitude = double_latitude, longitude = double_longitude, altitude = double_altitude, accuracy)
+    timezone <- names(all_timezones)[as.vector(all_timezones)==max(all_timezones)]
+    outputMobility <- MobilityFeatures(location, ACCURACY_LIM = accuracy_limit, tz = timezone)
+    
+    if(is.null(outputMobility)){
+      location_features <- create_empty_file()
+    } else {
+      # Copy index (dates) as a column 
+      features <- cbind(rownames(outputMobility$featavg), outputMobility$featavg)
+      features <- as.data.frame(features)
+      features[-1] <- lapply(lapply(features[-1], as.character), as.numeric)
+      colnames(features)=c("local_date",tolower(colnames(outputMobility$featavg)))
+      location_features <- left_join(features, location_minutes_used, by = "local_date")
+    }
+    
+  }
+  write.csv(location_features, snakemake@output[[1]], row.names =FALSE)
+}
+
+barnett_daily_features(snakemake)

src/features/phone_locations/barnett/main.R

@@ -1,10 +1,8 @@
 source("renv/activate.R")
 library("dplyr", warn.conflicts = F)
 library("stringr")
-
+library("lubridate")
-# Load Ian Barnett's code. Taken from https://scholar.harvard.edu/ibarnett/software/gpsmobility
+library("purrr")
-file.sources = list.files(c("src/features/phone_locations/barnett/library"), pattern="*.R$", full.names=TRUE, ignore.case=TRUE)
-sapply(file.sources,source,.GlobalEnv)
 
 create_empty_file <- function(requested_features){
   return(data.frame(local_segment= character(), 
@@ -27,82 +25,59 @@ create_empty_file <- function(requested_features){
   ) %>% select(all_of(requested_features)))
 }
 
+summarise_multiday_segments <- function(segments, features){
+  features <- features %>% mutate(local_date=ymd(local_date))
+  segments <- segments %>% extract(col = local_segment, 
+                                   into = c ("local_segment_start_datetime", "local_segment_end_datetime"),
+                                   ".*#(.*) .*,(.*) .*", 
+                                   remove = FALSE) %>% 
+    mutate(local_segment_start_datetime = ymd(local_segment_start_datetime),
+           local_segment_end_datetime = ymd(local_segment_end_datetime)) %>% 
+    group_by(local_segment) %>% 
+    nest() %>% 
+    mutate(data = map(data, function(nested_data, nested_features){
+      
+      summary <- nested_features %>% filter(local_date >= nested_data$local_segment_start_datetime &
+                                          local_date <= nested_data$local_segment_end_datetime)
+      if(nrow(summary) > 0)
+        summary <-  summary %>% 
+                 summarise(across(c(hometime, disttravelled, siglocsvisited, minutes_data_used), sum),
+                           across(c(maxdiam, maxhomedist), max),
+                           across(c(rog, avgflightlen, stdflightlen, avgflightdur, stdflightdur, probpause, siglocentropy, circdnrtn, wkenddayrtn, minsmissing), mean))
+      return(summary)
+      
+    }, features)) %>% 
+    unnest(cols = everything()) %>%
+    ungroup()
+  return(segments)
+}
+
 barnett_features <- function(sensor_data_files, time_segment, params){
-  
-  location_data <-  read.csv(sensor_data_files[["sensor_data"]], stringsAsFactors = FALSE)
   location_features <- NULL
-
+  daily_features <- read.csv(sensor_data_files[["barnett_daily"]], stringsAsFactors = FALSE)
-  location <- location_data
+  location <- read.csv(sensor_data_files[["sensor_data"]], stringsAsFactors = FALSE)
-  accuracy_limit <- params[["ACCURACY_LIMIT"]]
   minutes_data_used <- params[["MINUTES_DATA_USED"]]
-
-  # Compute what features were requested
   available_features <- c("hometime","disttravelled","rog","maxdiam", "maxhomedist","siglocsvisited","avgflightlen", "stdflightlen",
-                          "avgflightdur","stdflightdur", "probpause","siglocentropy","minsmissing", "circdnrtn","wkenddayrtn")
+                          "avgflightdur","stdflightdur", "probpause","siglocentropy", "circdnrtn","wkenddayrtn")
   requested_features <- intersect(unlist(params["FEATURES"], use.names = F), available_features)
   requested_features <- c("local_segment", requested_features)
   if(minutes_data_used)
     requested_features <- c(requested_features, "minutes_data_used")
   
-  # Excludes datasets with less than 24 hours of data
+  if (nrow(location) > 0 & nrow(daily_features) > 0){
-  if(max(location$timestamp) - min(location$timestamp) < 86400000)
-    location <- head(location, 0)
-
-  if (nrow(location) > 1){
-    # Filter by segment and skipping any non-daily segment
     location <- location %>% filter_data_by_segment(time_segment)
-    
     datetime_start_regex = "[0-9]{4}[\\-|\\/][0-9]{2}[\\-|\\/][0-9]{2} 00:00:00"
     datetime_end_regex = "[0-9]{4}[\\-|\\/][0-9]{2}[\\-|\\/][0-9]{2} 23:59:59"
     location <- location %>% mutate(is_daily = str_detect(local_segment, paste0(time_segment, "#", datetime_start_regex, ",", datetime_end_regex))) 
-
+    if(nrow(location) == 0 || !all(location$is_daily)){
-    if(!all(location$is_daily)){
+      message(paste("Barnett's location features cannot be computed for data or time segmentes that do not span entire days (00:00:00 to 23:59:59). Skipping ", time_segment))
-      message(paste("Barnett's location features cannot be computed for time segmentes that are not daily (cover 00:00:00 to 23:59:59 of every day). Skipping ", time_segment))
       location_features <- create_empty_file(requested_features)  
     } else {
-      # Count how many minutes of data we use to get location features
+      location_dates_segments <- location %>% select(local_segment) %>% distinct(local_segment, .keep_all = TRUE)
-      # Some minutes have multiple fused rows
+      features <- summarise_multiday_segments(location_dates_segments, daily_features)
-      location_minutes_used <- location %>% 
-        group_by(local_date, local_hour) %>% 
-        summarise(n_minutes = n_distinct(local_minute), .groups = 'drop_last') %>% 
-        group_by(local_date) %>% 
-        summarise(minutes_data_used = sum(n_minutes), .groups = 'drop_last') %>% 
-        select(local_date, minutes_data_used)
-
-      # Save time segment to attach it later
-      location_dates_segments <- location %>% select(local_date, local_segment) %>% distinct(local_date, .keep_all = TRUE)
-
-      # Select only the columns that the algorithm needs
-      all_timezones <- table(location %>% pull(local_timezone))
-      location <- location %>% select(timestamp, latitude = double_latitude, longitude = double_longitude, altitude = double_altitude, accuracy)
-      if(nrow(location %>% filter(accuracy < accuracy_limit)) > 1){
-        timezone <- names(all_timezones)[as.vector(all_timezones)==max(all_timezones)]
-        outputMobility <- MobilityFeatures(location, ACCURACY_LIM = accuracy_limit, tz = timezone)
-      } else {
-        print(paste("Cannot compute Barnett location features because there are no rows with an accuracy value lower than ACCURACY_LIMIT", accuracy_limit))
-        outputMobility <- NULL
-      }
-
-      if(is.null(outputMobility)){
-        location_features <- create_empty_file(requested_features)
-      } else{
-        # Copy index (dates) as a column 
-        features <- cbind(rownames(outputMobility$featavg), outputMobility$featavg)
-        features <- as.data.frame(features)
-        features[-1] <- lapply(lapply(features[-1], as.character), as.numeric)
-        colnames(features)=c("local_date",tolower(colnames(outputMobility$featavg)))
-        # Add the minute count column
-        features <- left_join(features, location_minutes_used, by = "local_date")
-        # Add the time segment column for consistency
-        features <- left_join(features, location_dates_segments, by = "local_date")
       location_features <- features %>% select(all_of(requested_features))
     } 
-    } 
+  } else
-  } else {
     location_features <- create_empty_file(requested_features)  
-  }
-
-  if(ncol(location_features) != length(requested_features))
-      stop(paste0("The number of features in the output dataframe (=", ncol(location_features),") does not match the expected value (=", length(requested_features),"). Verify your barnett location features"))
   return(location_features)
 }