nimare.meta.cbmr.CBMRInference

class CBMRInference(device='cpu')

Bases: object

Statistical inference on fitted CBMR results.

Notes

The CBMR API design now centers inference on CBMRResult. This class remains the lower-level implementation used by those result helpers and by advanced users who want to call fit() and transform() directly.

Added in version 0.1.0.

Parameters:

device (string, optional) – Device type (‘cpu’ or ‘cuda’) represents the device on which operations will be allocated. Default is ‘cpu’.

Methods

create_contrast(contrast_name[, source])	Create contrast matrix for generalized hypothesis testing (GLH).
create_regular_expressions()	Create regular expressions for parsing contrast names.
display()	Display Groups and Moderator names and order.
fit(result)	Fit CBMRInference instance.
fit_transform(result[, t_con_groups, ...])	Fit and transform.
transform([t_con_groups, t_con_moderators])	Conduct generalized linear hypothesis (GLH) testing on CBMR estimates.

create_contrast(contrast_name, source='groups')

Create contrast matrix for generalized hypothesis testing (GLH).

Named group contrasts may refer to a single group (for a homogeneity test) or a pairwise comparison such as group_a-group_b. Named moderator contrasts follow the same pattern.

Parameters:

• contrast_name (string or sequence of string) – Name or names of the contrasts to construct.

• source ({"groups", "moderators"}, optional) – Whether to build group or moderator contrasts.

create_regular_expressions()

Create regular expressions for parsing contrast names.

creates the following attributes: self.groups_regular_expression: regular expression for parsing group names self.moderators_regular_expression: regular expression for parsing moderator names

usage: >>> self.groups_regular_expression.match(“group1 - group2”).groupdict()

display()

Display Groups and Moderator names and order.

fit(result)

Fit CBMRInference instance.

Parameters:

result (CBMRResult) – Fitted CBMR result containing regression coefficient tables and spatial intensity maps.

fit_transform(result, t_con_groups=None, t_con_moderators=None)

Fit and transform.

transform(t_con_groups=None, t_con_moderators=None)

Conduct generalized linear hypothesis (GLH) testing on CBMR estimates.

Estimate group-wise spatial regression coefficients and its standard error via inverse Fisher Information matrix, estimate standard error of group-wise log intensity, group-wise intensity via delta method. For NB or clustered model, estimate regression coefficient of overdispersion. Similarly, estimate regression coefficient of experiment-level moderators (if exist), as well as its standard error via Fisher Information matrix. Save these outcomes in tables. Also, estimate group-wise spatial intensity (per experiment) and save the results in maps.

Parameters:

• t_con_groups (bool, dict, list, tuple, str, or None, optional) – Group inference specification. Use None or True to test all groups, False to skip group inference, named contrasts such as "group_a-group_b" or ("group_a", "group_b") for pairwise tests, a dict mapping names to contrast arrays, or raw contrast arrays.

• t_con_moderators (bool, dict, list, tuple, str, or None, optional) – Moderator inference specification with the same accepted forms as t_con_groups.