I0617 rescu sensors eval

volumes/rescu/README.md

@@ -1,5 +1,5 @@
 
-# *The RESCU schema is now deprecated. Please refer to [`vds` schema](../vds/readme.md).*
+# *The RESCU schema is now deprecated. Please refer to [`vds` schema](../vds/readme.md).* <!-- omit in toc -->
 - Use `vds.counts_15min` instead of `rescu.volumes_15min`
 - Identify RESCU sensors in the new schema using `vds.detector_inventory WHERE det_type = 'RESCU Detectors'`
 

volumes/vds/readme.md

@@ -23,6 +23,8 @@
     - [vds.counts\_15min\_bylane](#vdscounts_15min_bylane)
     - [vds.veh\_speeds\_15min](#vdsveh_speeds_15min)
     - [vds.veh\_length\_15min](#vdsveh_length_15min)
+    - [vds.individual\_outages](#vdsindividual_outages)
+    - [vds.network\_outages](#vdsnetwork_outages)
   - [Raw Data](#raw-data)
     - [vds.raw\_vdsdata](#vdsraw_vdsdata)
     - [vds.raw\_vdsvehicledata](#vdsraw_vdsvehicledata)
@@ -142,8 +144,8 @@ The regular detectors (DET) may have some utility but it is hard to tell with th
 - In some cases you may find it easier to select from only the partition of interest. eg. `FROM vds.raw_vdsdata_div2_202308` instead of more verbose ```FROM vds.raw_vdsdata WHERE division_id = 2 and dt >= '2023-08-01 00:00:00'::timestamp....```.
 - For RESCU requests, make use of the manually defined fields in `vds.detector_inventory` for easy filtering.  
 - Data quality checks have not been implemented in this new schema. For examples of past work see:  
-      - @scann0n did some work on identifying good date ranges for RESCU sensors based on volumes from days with all 96 15-minute bins present which is written up [here](https://github.com/CityofToronto/bdit_data-sources/blob/master/volumes/rescu/README.md#6--how-often-are-data-quality-assessment-processes-for-the-data-undertaken).
-      - @gabrielwol did work to identify periods of network wide or individual sensor outages on the RESCU network which is written up [here](https://github.com/CityofToronto/bdit_data-sources/blob/3ba3af5068e96191caffab524d42ae52fe7be7b2/volumes/rescu/README.md#1--are-there-known-data-gapsincomplete-data)  
+  - @scann0n did some work on identifying good date ranges for RESCU sensors based on volumes from days with all 96 15-minute bins present which is written up [here](../rescu/README.md#6--how-often-are-data-quality-assessment-processes-for-the-data-undertaken).  
+  - @gabrielwol created some helpful views to identify periods of [network wide](#vdsnetwork_outages) or [individual sensor](#vdsindividual_outages) outages on the RESCU network.  
 
 ## Lookup Tables and Views
 
@@ -394,6 +396,39 @@ Row count: 4,622,437
 
 </div>
 
+### vds.individual_outages
+
+A view of individual sensor outages of at least 30 minutes. Could be used in looking for days without outages of a certain duration, or in identifying unreliable sensors with multiple smaller outages. 
+
+It is recommended to add a where clause using `vdsconfig_uid` to get faster results for a sensor. 
+
+| column_name    | data_type | sample |
+|:---------------|:----------|:-------|
+| vdsconfig_uid | integer  | 516 |
+| entity_location_uid | integer  | 10238 |
+| division_id | smallint  | 2  |
+| time_start |  timestamp without time zone    | "2004-02-26 00:15:00" |
+| time_end |  timestamp without time zone    | "2004-02-27 00:00:00" |
+| date_start |  date    | "2004-02-26" |
+| date_end |  date    | "2004-02-27" |
+| time_range |  tsrange    | "[""2004-02-26 00:15:00"",""2004-02-27 00:00:00""]" |
+| date_range |  daterange    | "[2004-02-26,2004-02-28)" |
+| duration_days |  numeric     | 1.0  |
+
+### vds.network_outages
+
+`network_outages` view can be used to identify/eliminate dates from a study where all detectors are inactive. Runs in about 20s for entire network. Can be used for network health / uptime calculations.  
+
+| column_name    | data_type | sample |
+|:---------------|:----------|:-------|
+| time_start    |  timestamp without time zone    | "1993-01-11 17:00:00"  |
+| time_end    |  timestamp without time zone    | "1993-01-14 06:45:00"  |
+| date_start    |  date    | 1993-01-11  |
+| date_end    |  date    | 1993-01-14  |
+| time_range    |  tsrange    | ["1993-01-11 17:00:00","1993-01-14 06:45:00"]  |
+| date_range    |  daterange    | [1993-01-11,1993-01-15)  |
+| duration_days    |  numeric     | 2.5729166666666667  |
+
 ## Raw Data
 
 ### vds.raw_vdsdata

volumes/vds/sql/views/create-view-individual-outages.sql

@@ -0,0 +1,64 @@
+CREATE OR REPLACE VIEW vds.individual_outages
+AS
+
+-- make a table of outages by rescu detector.
+-- Outages includes network wide outages. Couldn't resolve overlapping individual and network
+-- wide outages using this method (no longer cross joining timebins and detectors).
+-- could be possible using future range operator: anymultirange - anymultirange → anymultirange
+-- runs in 2 minutes for all of rescu.volumes15min
+-- doesn't include outages that end after last data point (by detector)
+
+WITH bin_gaps AS (
+    SELECT
+        vdsconfig_uid,
+        entity_location_uid,
+        division_id,
+        datetime_15min,
+        count_15min,
+        LAG(datetime_15min, 1) OVER sensor + interval '15 minutes' AS gap_start,
+        datetime_15min - interval '15 minutes' AS gap_end,
+        CASE datetime_15min - LAG(datetime_15min, 1) OVER sensor
+            WHEN '00:15:00' THEN 0
+            ELSE 1
+        END AS bin_break, --identify non-consecutive bins
+        datetime_15min - LAG(datetime_15min, 1) OVER sensor AS bin_gap --duration of gap
+    FROM vds.counts_15min_div2
+    WHERE COALESCE(count_15min, 0) > 0
+    WINDOW sensor AS (
+        PARTITION BY vdsconfig_uid, entity_location_uid, division_id ORDER BY datetime_15min
+    )
+)
+
+SELECT
+    bt.vdsconfig_uid,
+    bt.entity_location_uid,
+    bt.division_id,
+    bt.gap_start AS time_start,
+    bt.gap_end AS time_end,
+    (bt.gap_start)::date AS date_start,
+    (bt.gap_end)::date AS date_end,
+    tsrange(bt.gap_start, bt.gap_end, '[]') AS time_range,
+    daterange((bt.gap_start)::date, (bt.gap_end)::date, '[]') AS date_range,
+    EXTRACT(
+        EPOCH FROM (bt.gap_end + interval '15 minutes') - bt.gap_start
+    ) / 86400 AS duration_days
+FROM bin_gaps AS bt
+WHERE
+    bt.bin_break = 1
+    AND bt.bin_gap IS NOT NULL
+    --Start and end are inclusive so interval 15 minutes implies a gap of 2 bins or 30 minutes
+    AND bt.gap_end - bt.gap_start >= interval '15 minutes'
+ORDER BY
+    bt.vdsconfig_uid,
+    bt.entity_location_uid,
+    bt.division_id,
+    bt.gap_start;
+
+ALTER TABLE vds.individual_outages OWNER TO vds_admins;
+
+GRANT SELECT ON TABLE vds.individual_outages TO bdit_humans;
+
+COMMENT ON VIEW vds.individual_outages IS 
+'''A view which identifies individual sensor outages in the VDS network.
+Note: does not indclude ongoing outages. See vds.detector_inventory.last_active for that purpose.
+Runs in around 1:30 for entire dataset or include a WHERE filter on vdsconfig_uid for speed.''';

volumes/vds/sql/views/create-view-network-outages.sql

@@ -0,0 +1,56 @@
+--network outages using partition, lag (sare's method).
+--much faster, runs in 1s for all of vds. 
+CREATE OR REPLACE VIEW vds.network_outages AS
+
+WITH rescu_summary AS (
+    SELECT
+        datetime_15min AS datetime_bin,
+        SUM(count_15min) AS volume
+    FROM vds.counts_15min_div2
+    GROUP BY datetime_15min
+    HAVING COALESCE(SUM(count_15min), 0) > 0
+
+    UNION
+
+    SELECT --quick way to include ongoing outage
+        now()::date AS datetime_bin,
+        1 AS volume
+    ORDER BY datetime_bin
+),
+
+bin_gaps AS (
+    SELECT
+        datetime_bin,
+        volume,
+        datetime_bin - interval '15 minutes' AS gap_end,
+        CASE datetime_bin - LAG(datetime_bin, 1) OVER (ORDER BY datetime_bin)
+            WHEN '00:15:00' THEN 0
+            ELSE 1
+        END AS bin_break,
+        datetime_bin - LAG(datetime_bin, 1) OVER (ORDER BY datetime_bin) AS bin_gap,
+        LAG(datetime_bin, 1) OVER (ORDER BY datetime_bin) + interval '15 minutes' AS gap_start
+    FROM rescu_summary
+)
+
+-- calculate the start and end times of gaps that are longer than 15 minutes
+SELECT
+    gap_start AS time_start,
+    gap_end AS time_end,
+    gap_start::date AS date_start,
+    gap_end::date AS date_end,
+    tsrange(gap_start, gap_end, '[]') AS time_range,
+    daterange(gap_start::date, gap_end::date, '[]') AS date_range,
+    --no minimum duration for a network wide outage
+    EXTRACT(EPOCH FROM gap_end - gap_start) / 86400 AS duration_days
+FROM bin_gaps
+WHERE
+    bin_break = 1
+    AND bin_gap IS NOT NULL
+ORDER BY gap_start;
+
+ALTER TABLE vds.network_outages OWNER TO vds_admins;
+
+GRANT SELECT ON TABLE vds.network_outages TO bdit_humans;
+
+COMMENT ON VIEW vds.network_outages
+IS 'A view which identifies network wide outages in the VDS network.';