Extrapolate travel time across missing data

[Nate-Wessel]

@chmnata can you confirm that the SQL in this block matches your understanding of how we are currently measuring average travel time using the congestion network?

bdit_tt_request_app/backend/app/get_travel_time.py

Lines 18 to 51 in 56b1b34

	-- Aggregate segments to corridor on a daily, hourly basis
	WITH corridor_hourly_daily_agg AS (
	SELECT
	cn.dt,
	cn.hr,
	SUM(cn.tt) AS corr_hourly_daily_tt
	FROM congestion.network_segments_daily AS cn
	WHERE
	cn.segment_id::integer = ANY(%(seglist)s)
	AND cn.hr <@ %(time_range)s::numrange
	AND date_part('ISODOW', cn.dt)::integer = ANY(%(dow_list)s)
	AND cn.dt <@ %(date_range)s::daterange
	{tt_holiday_clause}
	GROUP BY
	cn.dt,
	cn.hr
	-- where corridor has at least 80pct of links with data
	HAVING SUM(cn.length_w_data) >= %(length_m)s::numeric * 0.8
	),
	-- Average the hours selected into daily period level data
	corridor_period_daily_avg_tt AS (
	SELECT
	dt,
	AVG(corr_hourly_daily_tt) AS avg_corr_period_daily_tt
	FROM corridor_hourly_daily_agg
	GROUP BY
	dt
	)
	-- Average all the days with data to get period level data for each date range
	SELECT
	ROUND(AVG(avg_corr_period_daily_tt) / 60, 2) AS average_tt_min
	FROM corridor_period_daily_avg_tt

[chmnata]

If we are using link->segment-> corridor with adjusted tt aka tt, in the where clause for agg_tt_query, we should add a is_valid = TRUE. The adjusted TT column is calculated for all segment even if the length w data is below 80%. The is_valid column is true when there are 80% or more links with data in that segment.
And then instead of using length_w_data as a base for the 80% check, we should use the sum length of the segments with is_valid = true to check against the corridor length.

If we are using unadjusted_tt, we will just need to adjust the sum(unadjusted_tt) with corridor_length/sum link_length, on top of the previous query used.

[Nate-Wessel]

If we are using unadjusted_tt, we will just need to adjust the sum(unadjusted_tt) with corridor_length/sum link_length, on top of the previous query used.

This approach seems more robust as it would extrapolate across any missing segments. Can you take another look now?

[Nate-Wessel]

Results seem reasonable. Slightly higher travel times than before.