编辑

通过

你当前正在访问 Microsoft Azure Global Edition 技术文档网站。 如果需要访问由世纪互联运营的 Microsoft Azure 中国技术文档网站，请访问 https://docs.azure.cn。

Tutorial: Use aggregation functions

  • 项目

Applies to: ✅ Microsoft Fabric ✅ Azure Data Explorer ✅ Azure Monitor ✅ Microsoft Sentinel

Aggregation functions allow you to group and combine data from multiple rows into a summary value. The summary value depends on the chosen function, for example a count, maximum, or average value.

In this tutorial, you'll learn how to:

The examples in this tutorial use the StormEvents table, which is publicly available in the help cluster. To explore with your own data, create your own free cluster.

The examples in this tutorial use the StormEvents table, which is publicly available in the Weather analytics sample data.

This tutorial builds on the foundation from the first tutorial, Learn common operators.

Prerequisites

To run the following queries, you need a query environment with access to the sample data. You can use one of the following:

  • A Microsoft account or Microsoft Entra user identity to sign in to the help cluster

Use the summarize operator

The summarize operator is essential to performing aggregations over your data. The summarize operator groups together rows based on the by clause and then uses the provided aggregation function to combine each group in a single row.

Find the number of events by state using summarize with the count aggregation function.

StormEvents
| summarize TotalStorms = count() by State

Output

State TotalStorms
TEXAS 4701
KANSAS 3166
IOWA 2337
ILLINOIS 2022
MISSOURI 2016
... ...

Visualize query results

Visualizing query results in a chart or graph can help you identify patterns, trends, and outliers in your data. You can do this with the render operator.

Throughout the tutorial, you'll see examples of how to use render to display your results. For now, let's use render to see the results from the previous query in a bar chart.

StormEvents
| summarize TotalStorms = count() by State
| render barchart

Screenshot of total storms by state bar chart created with the render operator.

Conditionally count rows

When analyzing your data, use countif() to count rows based on a specific condition to understand how many rows meet the given criteria.

The following query uses countif() to count of storms that caused damage. The query then uses the top operator to filter the results and display the states with the highest amount of crop damage caused by storms.

StormEvents
| summarize StormsWithCropDamage = countif(DamageCrops > 0) by State
| top 5 by StormsWithCropDamage

Output

State StormsWithCropDamage
IOWA 359
NEBRASKA 201
MISSISSIPPI 105
NORTH CAROLINA 82
MISSOURI 78

Group data into bins

To aggregate by numeric or time values, you'll first want to group the data into bins using the bin() function. Using bin() can help you understand how values are distributed within a certain range and make comparisons between different periods.

The following query counts the number of storms that caused crop damage for each week in 2007. The 7d argument represents a week, as the function requires a valid timespan value.

StormEvents
| where StartTime between (datetime(2007-01-01) .. datetime(2007-12-31)) 
    and DamageCrops > 0
| summarize EventCount = count() by bin(StartTime, 7d)

Output

StartTime EventCount
2007-01-01T00:00:00Z 16
2007-01-08T00:00:00Z 20
2007-01-29T00:00:00Z 8
2007-02-05T00:00:00Z 1
2007-02-12T00:00:00Z 3
... ...

Add | render timechart to the end of the query to visualize the results.

Screenshot of the crop damage by week time chart rendered by the previous query.

Note

bin() is similar to the floor() function in other programming languages. It reduces every value to the nearest multiple of the modulus that you supply and allows summarize to assign the rows to groups.

Calculate the min, max, avg, and sum

To learn more about types of storms that cause crop damage, calculate the min(), max(), and avg() crop damage for each event type, and then sort the result by the average damage.

Note that you can use multiple aggregation functions in a single summarize operator to produce several computed columns.

StormEvents
| where DamageCrops > 0
| summarize
    MaxCropDamage=max(DamageCrops), 
    MinCropDamage=min(DamageCrops), 
    AvgCropDamage=avg(DamageCrops)
    by EventType
| sort by AvgCropDamage

Output

EventType MaxCropDamage MinCropDamage AvgCropDamage
Frost/Freeze 568600000 3000 9106087.5954198465
Wildfire 21000000 10000 7268333.333333333
Drought 700000000 2000 6763977.8761061952
Flood 500000000 1000 4844925.23364486
Thunderstorm Wind 22000000 100 920328.36538461538
... ... ... ...

The results of the previous query indicate that Frost/Freeze events resulted in the most crop damage on average. However, the bin() query showed that events with crop damage mostly took place in the summer months.

Use sum() to check the total number of damaged crops instead of the amount of events that caused some damage, as done with count() in the previous bin() query.

StormEvents
| where StartTime between (datetime(2007-01-01) .. datetime(2007-12-31)) 
    and DamageCrops > 0
| summarize CropDamage = sum(DamageCrops) by bin(StartTime, 7d)
| render timechart

Screenshot of time chart showing crop damage by week.

Now you can see a peak in crop damage in January, which probably was due to Frost/Freeze.

Tip

Use minif(), maxif(), avgif(), and sumif() to perform conditional aggregations, like we did when in the conditionally count rows section.

Calculate percentages

Calculating percentages can help you understand the distribution and proportion of different values within your data. This section covers two common methods for calculating percentages with the Kusto Query Language (KQL).

Calculate percentage based on two columns

Use count() and countif to find the percentage of storm events that caused crop damage in each state. First, count the total number of storms in each state. Then, count the number of storms that caused crop damage in each state.

Then, use extend to calculate the percentage between the two columns by dividing the number of storms with crop damage by the total number of storms and multiplying by 100.

To ensure that you get a decimal result, use the todouble() function to convert at least one of the integer count values to a double before performing the division.

StormEvents
| summarize 
    TotalStormsInState = count(),
    StormsWithCropDamage = countif(DamageCrops > 0)
    by State
| extend PercentWithCropDamage = 
    round((todouble(StormsWithCropDamage) / TotalStormsInState * 100), 2)
| sort by StormsWithCropDamage

Output

State TotalStormsInState StormsWithCropDamage PercentWithCropDamage
IOWA 2337 359 15.36
NEBRASKA 1766 201 11.38
MISSISSIPPI 1218 105 8.62
NORTH CAROLINA 1721 82 4.76
MISSOURI 2016 78 3.87
... ... ... ...

Note

When calculating percentages, convert at least one of the integer values in the division with todouble() or toreal(). This will ensure that you don't get truncated results due to integer division. For more information, see Type rules for arithmetic operations.

Calculate percentage based on table size

To compare the number of storms by event type to the total number of storms in the database, first save the total number of storms in the database as a variable. Let statements are used to define variables within a query.

Since tabular expression statements return tabular results, use the toscalar() function to convert the tabular result of the count() function to a scalar value. Then, the numeric value can be used in the percentage calculation.

let TotalStorms = toscalar(StormEvents | summarize count());
StormEvents
| summarize EventCount = count() by EventType
| project EventType, EventCount, Percentage = todouble(EventCount) / TotalStorms * 100.0

Output

EventType EventCount Percentage
Thunderstorm Wind 13015 22.034673077574237
Hail 12711 21.519994582331627
Flash Flood 3688 6.2438627975485055
Drought 3616 6.1219652592015716
Winter Weather 3349 5.669928554498358
... ... ...

Extract unique values

Use make_set() to turn a selection of rows in a table into an array of unique values.

The following query uses make_set() to create an array of the event types that cause deaths in each state. The resulting table is then sorted by the number of storm types in each array.

StormEvents
| where DeathsDirect > 0 or DeathsIndirect > 0
| summarize StormTypesWithDeaths = make_set(EventType) by State
| project State, StormTypesWithDeaths
| sort by array_length(StormTypesWithDeaths)

Output

State StormTypesWithDeaths
CALIFORNIA ["Thunderstorm Wind","High Surf","Cold/Wind Chill","Strong Wind","Rip Current","Heat","Excessive Heat","Wildfire","Dust Storm","Astronomical Low Tide","Dense Fog","Winter Weather"]
TEXAS ["Flash Flood","Thunderstorm Wind","Tornado","Lightning","Flood","Ice Storm","Winter Weather","Rip Current","Excessive Heat","Dense Fog","Hurricane (Typhoon)","Cold/Wind Chill"]
OKLAHOMA ["Flash Flood","Tornado","Cold/Wind Chill","Winter Storm","Heavy Snow","Excessive Heat","Heat","Ice Storm","Winter Weather","Dense Fog"]
NEW YORK ["Flood","Lightning","Thunderstorm Wind","Flash Flood","Winter Weather","Ice Storm","Extreme Cold/Wind Chill","Winter Storm","Heavy Snow"]
KANSAS ["Thunderstorm Wind","Heavy Rain","Tornado","Flood","Flash Flood","Lightning","Heavy Snow","Winter Weather","Blizzard"]
... ...

Bucket data by condition

The case() function groups data into buckets based on specified conditions. The function returns the corresponding result expression for the first satisfied predicate, or the final else expression if none of the predicates are satisfied.

This example groups states based on the number of storm-related injuries their citizens sustained.

StormEvents
| summarize InjuriesCount = sum(InjuriesDirect) by State
| extend InjuriesBucket = case (
                              InjuriesCount > 50,
                              "Large",
                              InjuriesCount > 10,
                              "Medium",
                              InjuriesCount > 0,
                              "Small",
                              "No injuries"
                          )
| sort by State asc

Output

State InjuriesCount InjuriesBucket
ALABAMA 494 Large
ALASKA 0 No injuries
AMERICAN SAMOA 0 No injuries
ARIZONA 6 Small
ARKANSAS 54 Large
ATLANTIC NORTH 15 Medium
... ... ...

Create a pie chart to visualize the proportion of states that experienced storms resulting in a large, medium, or small number of injuries.

StormEvents
| summarize InjuriesCount = sum(InjuriesDirect) by State
| extend InjuriesBucket = case (
                              InjuriesCount > 50,
                              "Large",
                              InjuriesCount > 10,
                              "Medium",
                              InjuriesCount > 0,
                              "Small",
                              "No injuries"
                          )
| summarize InjuryBucketByState=count() by InjuriesBucket
| render piechart

Screenshot of web UI pie chart rendered by the previous query.

Perform aggregations over a sliding window

The following example shows how to summarize columns using a sliding window.

The query calculates the minimum, maximum, and average property damage of tornados, floods, and wildfires using a sliding window of seven days. Each record in the result set aggregates the preceding seven days, and the results contain a record per day in the analysis period.

Here's a step-by-step explanation of the query:

  1. Bin each record to a single day relative to windowStart.
  2. Add seven days to the bin value to set the end of the range for each record. If the value is out of the range of windowStart and windowEnd, adjust the value accordingly.
  3. Create an array of seven days for each record, starting from the current day of the record.
  4. Expand the array from step 3 with mv-expand in order to duplicate each record to seven records with one-day intervals between them.
  5. Perform the aggregations for each day. Due to step 4, this step actually summarizes the previous seven days.
  6. Exclude the first seven days from the final result because there's no seven-day lookback period for them.
let windowStart = datetime(2007-07-01);
let windowEnd = windowStart + 13d;
StormEvents
| where EventType in ("Tornado", "Flood", "Wildfire") 
| extend bin = bin_at(startofday(StartTime), 1d, windowStart) // 1
| extend endRange = iff(bin + 7d > windowEnd, windowEnd, 
                      iff(bin + 7d - 1d < windowStart, windowStart, 
                        iff(bin + 7d - 1d < bin, bin, bin + 7d - 1d))) // 2
| extend range = range(bin, endRange, 1d) // 3
| mv-expand range to typeof(datetime) // 4
| summarize min(DamageProperty), max(DamageProperty), round(avg(DamageProperty)) by Timestamp=bin_at(range, 1d, windowStart), EventType // 5
| where Timestamp >= windowStart + 7d; // 6

Output

The following result table is truncated. To see the full output, run the query.

Timestamp EventType min_DamageProperty max_DamageProperty avg_DamageProperty
2007-07-08T00:00:00Z Tornado 0 30000 6905
2007-07-08T00:00:00Z Flood 0 200000 9261
2007-07-08T00:00:00Z Wildfire 0 200000 14033
2007-07-09T00:00:00Z Tornado 0 100000 14783
2007-07-09T00:00:00Z Flood 0 200000 12529
2007-07-09T00:00:00Z Wildfire 0 200000 14033
2007-07-10T00:00:00Z Tornado 0 100000 31400
2007-07-10T00:00:00Z Flood 0 200000 12263
2007-07-10T00:00:00Z Wildfire 0 200000 11694
... ... ...

Next step

Now that you're familiar with common query operators and aggregation functions, go on to the next tutorial to learn how to join data from multiple tables.

Join data from multiple tables