Merge pull request #1080 from Algunenano/histogram_irq_improvement

Numeric histogram performance improvement
This commit is contained in:
Raúl Marín
2019-03-14 15:23:44 +01:00
committed by GitHub
3 changed files with 54 additions and 44 deletions
+1
View File
@@ -5,6 +5,7 @@ Released 2019-mm-dd
Announcements:
- Experimental support for listing features in a grid when the map uses the dynamic agregation.
- Numeric histogram performance improvement (#1080)
- Update deps:
- windshaft@4.13.3: Upgrade grainstore to version 1.11.0, do not hang when child process is not able to generate a Mapnik XML
@@ -4,50 +4,37 @@ const BaseHistogram = require('./base-histogram');
const debug = require('debug')('windshaft:dataview:numeric-histogram');
const utils = require('../../../utils/query-utils');
/** Query to get min and max values from the query */
/** Query to get min, max, count and (if necessary) bin number of the query */
const irqQueryTpl = ctx => `
__cdb_filtered_source AS (
SELECT *
FROM (${ctx.query}) __cdb_filtered_source_query
WHERE ${utils.handleFloatColumn(ctx)} IS NOT NULL
),
__cdb_basics AS (
__cdb_basics AS (
SELECT
*,
CASE
WHEN __cdb_total_rows = 0 OR __cdb_iqr = 0 THEN 1
ELSE GREATEST(
LEAST(
${ctx.minBins},
__cdb_total_rows::int),
LEAST(
${ctx.maxBins},
((__cdb_max_val - __cdb_min_val) / (2 * __cdb_iqr * power(__cdb_total_rows, 1/3)))::int)
)
END AS __cdb_bins_number
FROM
(
SELECT
max(${ctx.column}) AS __cdb_max_val,
min(${ctx.column}) AS __cdb_min_val,
count(1) AS __cdb_total_rows
FROM __cdb_filtered_source
)
`;
/* Query to calculate the number of bins (needs irqQueryTpl before it*/
const binsQueryTpl = ctx => `
__cdb_iqrange AS (
SELECT max(quartile_max) - min(quartile_max) AS __cdb_iqr
FROM (
SELECT quartile, max(_cdb_iqr_column) AS quartile_max from (
SELECT ${ctx.column} AS _cdb_iqr_column, ntile(4) over (order by ${ctx.column}
) AS quartile
FROM __cdb_filtered_source) _cdb_quartiles
WHERE quartile = 1 or quartile = 3
GROUP BY 1
) __cdb_iqr
),
__cdb_bins AS (
SELECT
CASE WHEN __cdb_total_rows = 0 OR __cdb_iqr = 0
THEN 1
ELSE GREATEST(
LEAST(${ctx.minBins}, CAST(__cdb_total_rows AS INT)),
LEAST(
CAST(((__cdb_max_val - __cdb_min_val) / (2 * __cdb_iqr * power(__cdb_total_rows, 1/3))) AS INT),
${ctx.maxBins}
)
)
END AS __cdb_bins_number
FROM __cdb_basics, __cdb_iqrange, __cdb_filtered_source
LIMIT 1
)
count(1) AS __cdb_total_rows,
${ctx.irq ? ctx.irq : `0`} AS __cdb_iqr
FROM
(
SELECT *
FROM (${ctx.query}) __cdb_filtered_source_query
WHERE ${utils.handleFloatColumn(ctx)} IS NOT NULL
) __cdb_filtered_source
) __cdb_basics_2
)
`;
const BIN_MIN_NUMBER = 6;
@@ -117,10 +104,12 @@ module.exports = class NumericHistogram extends BaseHistogram {
}
if (ctx.bins <= 0) {
ctx.bins = `__cdb_bins.__cdb_bins_number`;
extra_groupby += `, __cdb_bins.__cdb_bins_number`;
extra_tables += `, __cdb_bins`;
extra_queries = `WITH ${irqQueryTpl(ctx)}, ${binsQueryTpl(ctx)}`;
ctx.bins = `__cdb_basics.__cdb_bins_number`;
ctx.irq = `percentile_disc(0.75) within group (order by ${ctx.column})
- percentile_disc(0.25) within group (order by ${ctx.column})`;
extra_groupby += `, __cdb_basics.__cdb_bins_number`;
extra_tables = `, __cdb_basics`;
extra_queries = `WITH ${irqQueryTpl(ctx)}`;
}
return `
+20
View File
@@ -90,6 +90,26 @@ describe('histogram-dataview', function() {
});
});
it('should work with min >= start and max <= end, autodetect bins', function(done) {
var params = {
start: 50,
end: 500
};
this.testClient = new TestClient(mapConfig, 1234);
this.testClient.getDataview('pop_max_histogram', params, function(err, dataview) {
assert.ok(!err, err);
assert.ok(6 === dataview.bins_count, 'Unexpected bin count: ' + dataview.bins_count);
assert.ok(6 === dataview.bins.length, 'Unexpected number of bins: ' + dataview.bins.length);
dataview.bins.forEach(function(bin) {
assert.ok(bin.min >= params.start, 'bin min < start: ' + JSON.stringify(bin));
assert.ok(bin.max <= params.end, 'bin max > end: ' + JSON.stringify(bin));
});
done();
});
});
it('should get bin_width right when max > min in filter', function(done) {
var params = {
bins: 10,