Quantize (deprecated)

You can use the Quantize module to bin numbers.

Warning
This module is provided for backward compatibility with experiments created using the pre-release version of Azure Machine Learning, and will soon be deprecated. We recommend that you modify your experiments to use Quantize Data instead.

This module is useful for binning and flattening the distribution of continuous data

During binning, each input element is mapped to a bin by comparing its value against positions of bin edges. For example, if value is 1.5 and bin edges are 1,2,3, the element would be mapped to bin number 2. Value 0.5 would be mapped to bin number 1 (the underflow bin), and value 3.5 would be mapped to bin number 4 (the overflow bin).

The help for this module was based on an early release version and will not be updated because this module is deprecated. For up-to-date information about how quantization works, see the help topic for Quantize Data.

• The option Tag columns as categorical can be used to control whether the quantized columns become categorical variables.

• To apply different quantization rules to different columns, chain together multiple instances of the Quantize module, and in each instance select a subset of columns to quantize.

• The same binning rule is applied to all columns specified in list of columns to quantize. The output mode specifies how the quantized values are returned. Choices include appending input table, overwriting columns in input table, and returning result columns only.

• Input columns must be numeric, and for quantile binning there must be a sufficient range of data points to determine the quantiles. Otherwise an error or NaN result may occur.

The bin indices are 1-based. This is the natural convention for quantiles (1st quantile, 2nd quantile, and so on). The only exception is the case when the column to bin is sparse.

All NaNs and missing values are propagated from input to output column. The only exception is the case when the module returns quantile indexes. In this case all NaNs are promoted to missing values.

If the column to bin (quantize) is sparse, then the bin index offset (quantile offset) is used when resulting column is populated. The offset is chosen so that sparse 0 always goes to the bin with index 0 (quantile with value 0). As a result, sparse zeros are propagated from input to output column. Notice that processing of dense column always produces a result with a minimum bin index equal to 1 (in other words, the minimum quantile value equals the minimum value in the column). Processing of a sparse column produces a result with variable values for the minimum bin index (minimum quantile value).