Options for Data / Target drift
You can modify certain options when calculating the Data and Target drift.
Last updated
You can modify certain options when calculating the Data and Target drift.
Last updated
An example of setting custom options in the Data Drift report on California Housing Dataset:
You can set the custom options for the following Reports:
num_target_drift_tab (Numerical Target Drift)
cat_target_drift_tab ( Categorical Target Drift)
data_drift_tab (Data Drift)
You can specify the following parameters:
confidence: float or dict[str, float]. Default = 0.95.
Defines the confidence level for the statistical tests.
Applies to all features (if passed as float) or certain features (if passed as dictionary).
drift_share: float. Default = 0.5.
Sets the share of drifting features as a condition for Dataset Drift in the Data Drift report.
nbinsx: int or dict[str, int]. Default = 10.
Defines the number of bins in a histogram.
Applies to all features (if passed as int) or certain features (if passed as dictionary).
xbins: dict[str, int]. Default = None.
Defines the boundaries for the size of a specific bin in a histogram.
feature_stattest_func: Callable or Dict[str, Callable]. Default = None.
Defines a custom statistical test for drift detection in the Data Drift report.
Applies to all features (if passed as a _functio_n) or individual features (if a dictionary).
cat_target_stattest_func: Callable. Default = None.
Defines a custom statistical test to detect target drift in the Categorical Target Drift report.
num_target_stattest_func: Callable. Default = None.
Defines a custom statistical test to detect target drift in the Numerical Target Drift report.
Define a DataDriftOptions object. This is a single object for Data Drift and Target Drift Reports.
Note: when you pass the function as an argument it should satisfy two conditions:
takes as an argument two DataFrame columns (series) - reference and production data
returns a float - p_value
2. Pass it to the Dashboard class:
You can also set the options from the command-line interface. In this case, you cannot define the functions (e.g. change statistical tests).
The section below explains specific popular customizations in more detail.
You can override the default statistical tests that Evidently uses in the Data Drift report.
To do that, set the following option:
feature_stattest_func: Callable or Dict[str, Callable].
This option can take a function or a dictionary as an argument.
If you pass a function, this function will be used to detect drift for all features.
If you pass a dictionary, the custom functions will be used for the specified features. The default Evidently tests would apply to the rest.
To add an alternative test, you need to implement a function that would return a float (p-value) after receiving two DataFrame columns that correspond to the reference and current datasets.
We suggest using statistical tests from scipy or statsmodels or implementing your own.
Then, define the DataDriftOptions object as shown above.
The Data Drift report contains a component that confirms whether the drift was detected on the Dataset level.
To set custom drift conditions, you need to specify the following options:
“confidence” - statistical test confidence level (default value 0.95; float or dict)
“drift_share” - share of the drifted features (default value 0.5; float)
You can set the same confidence level for all features. In this case, specify a float value for the "confidence" option. The Dataset Drift will be detected if the “drift_share” share of the features drift at the defined “confidence” confidence level.
You can also set different confidence levels for different features. In this case, you should pass a dictionary for the "confidence" option. A custom confidence level will be applied for the specified features. The rest will have the default confidence level = 0.95.
Then, define the DataDriftOptions object as shown above.
You can customize how the distribution plots look for the individual features in the Data Drift report. It is helpful, for example, if you have NULL or other specific values and want to see them in a separate bin.
To customize the plots, specify the following options:
“nbinsx” - to set the number of bins (default value = 10, integer or dictionary)
If you pass an integer value, the selected number of bins will apply to all features.
If you pass a dictionary, then specified features will have a custom number of bins. The rest will have the default number of bins = 10.
“xbins” - to define the specific bin sizes (default value = none).
Dict("start"=value, "end"=value, "size"=value) or plotly.graph_objects.histogram.XBins
You can set different options for each feature. For example, you can specify “nbinsx” for one subset of the features, “xbins” for another, and apply defaults for the rest. Here is an example.
Once you specify the options, define the DataDriftOptions object as shown above.
The Data Drift report has two sets of histograms:
preview in the Data Drift table
an interactive plot inside the Data Drift table that expands when you click on each feature
Only “nbinsx”, if specified, impacts the histogram previews in the DataDrift table. In case you set both parameters, “xbins” will define the interactive plot, while “nbinsx” will affect the preview.
Both “nbinsx” and “xbins” can influence how the interactive plots look inside the table. If you set one parameter, it will define the plot view. If you set both parameters, “xbins” will have a priority
