the heatmap (even all the basic type vislib type visualizations) doesn't support to sort the x/y labels in the chart based on my understanding.
the terms in the buckets actually only reflect the es query string and to sort something in the es.
and actually what i want is something like this vega-vis heatmap. and this map is a real wafer map which is widely used in the semiconductor testing field, the x and y is the position and the cell color standards for the metric for this position.
I saw some similar topics here on sorting the labels or sorting the legends, and now I want to customize the kibana heatmap vis to have this sort by lable option. and actually I'm really new to the kibana development and know little about the vislib type visualizations. I'm not sure if some experts here can talk more about the details or share something on how to realize this feature?
yeah that behavior is indeed a bit misleading in the beginning. The reason why the y-axis is sorted other than expected, is these two aggregations are actually done in a nested way, i.e. you first aggregate those x-axis values (which thus have correct sorting) and then within each of the x-axis values, you do the aggregation for the y-axis. That way you will only achieve sorting within each of the x-axis slot, which you can actually also see in the chart, Each section on the y-axis in the following chart separated by lines is sorted in itself.
To achieve the chart you want, you should instead of a terms aggregation, rather use a Histogram aggregation on that field. Switch on Show empty buckets, and just specify the min max value you want for that chart to have. That way the y-axis should be sorted globally the way you want it to be.
Yes, thank you very much and your explanation is very clear. it does behaves LIKE the y-axis is sorted by the label while adding the empty cells into the row data series.
one thing is not so good is, e.g. it's not easy to judge some cells are EMPTY or just VALID data for the cells with value '0' in the centre part of the chart.
I think there is actually a couple of issues, that we could improve.
One is also allowing non nested aggregations for visualizing data, which is actually tracked via #1907. That way you wouldn't have that min/max issue since you could have two term aggregations on the same level.
Another suggestion would be, the ability to filter out labels/cell colors at the border of the chart (or that are 0. I don't think we have an issue for that yet, so feel free to open a feature request for it if relevant for you.
Btw: what actually happens if you leave the min/max value empty?
If only check the Show empty buckets and without the max and min settings, it wont's show the EMPTY data so the chart is still something like the Terms Sub Aggregation.
one question is you know if always contains the value '0', we can not take advantage of the Scale to Data Bounds options if the color range are all very centralized to a number.
I just noticed in kibana 6.2.5 the value of these EMPTY cell are not showed with the Show Labels checked. this is good but the color is set to Black if Scale to Data Bounds is checked. it's better if can use the White color for the EMPTY cells.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.
