Reidentifiability metric. This corresponds to a risk model similar to what is called “journalist risk” in the literature, except the attack dataset is statistically modeled instead of being perfectly known. This can be done using publicly available data (like the US Census), or using a custom statistical model (indicated as one or several BigQuery tables), or by extrapolating from the distribution of values in the input dataset. .. attribute:: quasi_ids
Required. Fields considered to be quasi-identifiers. No two columns can have the same tag.
Several auxiliary tables can be used in the analysis. Each custom_tag used to tag a quasi-identifiers column must appear in exactly one column of one auxiliary table.
Classes
AuxiliaryTable
An auxiliary table contains statistical information on the relative frequency of different quasi-identifiers values. It has one or several quasi-identifiers columns, and one column that indicates the relative frequency of each quasi-identifier tuple. If a tuple is present in the data but not in the auxiliary table, the corresponding relative frequency is assumed to be zero (and thus, the tuple is highly reidentifiable). .. attribute:: table
Required. Auxiliary table location.
Required. The relative frequency column must contain a floating-point number between 0 and 1 (inclusive). Null values are assumed to be zero.
TaggedField
A column with a semantic tag attached. .. attribute:: field
Required. Identifies the column.
A column can be tagged with a InfoType to use the relevant public dataset as a statistical model of population, if available. We currently support US ZIP codes, region codes, ages and genders. To programmatically obtain the list of supported InfoTypes, use ListInfoTypes with the supported_by=RISK_ANALYSIS filter.
If no semantic tag is indicated, we infer the statistical model from the distribution of values in the input data