datafu.pig.stats
Class Quantile

java.lang.Object
  org.apache.pig.EvalFunc<T>
      datafu.pig.util.SimpleEvalFunc<org.apache.pig.data.Tuple>
          datafu.pig.stats.Quantile

Direct Known Subclasses:

Median

public class Quantile
extends SimpleEvalFunc<org.apache.pig.data.Tuple>

Computes quantiles for a sorted input bag, using type R-2 estimation.

N.B., all the data is pushed to a single reducer per key, so make sure some partitioning is done (e.g., group by 'day') if the data is too large. That is, this isn't distributed quantiles.

Note that unlike datafu's StreamingQuantile algorithm, this implementation gives exact quantiles. But, it requires that the input bag to be sorted. Quantile must spill to disk when the input data is too large to fit in memory, which will contribute to longer runtimes. Because StreamingQuantile implements accumulate it can be much more efficient than Quantile for large input bags which do not fit well in memory.

The constructor takes a single integer argument that specifies the number of evenly-spaced quantiles to compute, e.g.,

• Quantile('3') yields the min, the median, and the max
• Quantile('5') yields the min, the 25th, 50th, 75th percentiles, and the max
• Quantile('101') yields the min, the max, and all 99 percentiles.

Alternatively the constructor can take the explicit list of quantiles to compute, e.g.

• Quantile('0.0','0.5','1.0') yields the min, the median, and the max
• Quantile('0.0','0.25','0.5','0.75','1.0') yields the min, the 25th, 50th, 75th percentiles, and the max

The list of quantiles need not span the entire range from 0.0 to 1.0, nor do they need to be evenly spaced, e.g.

• Quantile('0.5','0.90','0.95','0.99') yields the median, the 90th, 95th, and the 99th percentiles
• Quantile('0.0013','0.0228','0.1587','0.5','0.8413','0.9772','0.9987') yields the 0.13th, 2.28th, 15.87th, 50th, 84.13th, 97.72nd, and 99.87th percentiles

Example:

 define Quantile datafu.pig.stats.Quantile('0.0','0.5','1.0');

 -- input: 9,10,2,3,5,8,1,4,6,7
 input = LOAD 'input' AS (val:int);

 grouped = GROUP input ALL;

 -- produces: (1,5.5,10)
 quantiles = FOREACH grouped {
   sorted = ORDER input BY val;
   GENERATE Quantile(sorted);
 }
 

See Also:

Median, StreamingQuantile

Field Summary

Fields inherited from class org.apache.pig.EvalFunc

log, pigLogger, reporter, returnType

Constructor Summary

Quantile(java.lang.String... k)

Method Summary

org.apache.pig.data.Tuple	call(org.apache.pig.data.DataBag bag)
org.apache.pig.impl.logicalLayer.schema.Schema	outputSchema(org.apache.pig.impl.logicalLayer.schema.Schema input) Override outputSchema so we can verify the input schema at pig compile time, instead of runtime

Methods inherited from class datafu.pig.util.SimpleEvalFunc

exec, getReturnType

Methods inherited from class org.apache.pig.EvalFunc

finish, getArgToFuncMapping, getCacheFiles, getInputSchema, getLogger, getPigLogger, getReporter, getSchemaName, isAsynchronous, progress, setInputSchema, setPigLogger, setReporter, setUDFContextSignature, warn

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

Quantile

public Quantile(java.lang.String... k)

Method Detail

call

public org.apache.pig.data.Tuple call(org.apache.pig.data.DataBag bag)
                               throws java.io.IOException

Throws:

java.io.IOException

outputSchema

public org.apache.pig.impl.logicalLayer.schema.Schema outputSchema(org.apache.pig.impl.logicalLayer.schema.Schema input)

Description copied from class: SimpleEvalFunc

Override outputSchema so we can verify the input schema at pig compile time, instead of runtime

Overrides:

outputSchema in class SimpleEvalFunc<org.apache.pig.data.Tuple>

Parameters:

input - input schema

Returns:

call to super.outputSchema in case schema was defined elsewhere