Enum: DecompositionMethod

The algorithm that produced the components or latent variables, which decides what their values and signs mean. Independent of second_block, since one block structure is reachable by several algorithms: a brain-behaviour covariance can be decomposed by PLS or by CCA. Open vocabulary: write the source's own wording in the field when nothing here fits.

URI: parse:DecompositionMethod

Permissible Values

Value	Meaning	Description
ica	None	Independent component analysis
tensor_ica	None	ICA over a subject-by-time-by-space tensor, giving one shared spatial map wit...
joint_ica	None	ICA on features concatenated across modalities, sharing one mixing matrix
parallel_ica	None	ICA run on two blocks at once with their mixing profiles linked, so a compone...
pca	None	Principal component analysis, whose components are orthogonal rather than ind...
nmf	None	Non-negative matrix factorization, whose components cannot take negative valu...
dictionary_learning	None	A sparse decomposition learning an overcomplete set of components
pls	None	Partial least squares, where the source does not distinguish the correlation ...
pls_correlation	None	PLS correlation (PLS-C), decomposing the cross-block correlation matrix by SV...
pls_regression	None	PLS regression, which predicts one block from the other rather than decomposi...
cca	None	Canonical correlation analysis, which maximizes correlation between blocks wh...
mcca	None	Multi-set CCA, over three or more blocks

In Subsets

• mvp
• model_extracted

Identifier and Mapping Information

Schema Source

• from schema: https://neurostuff.org/schemas/neuroimaging-paper-parse

LinkML Source

name: DecompositionMethod
description: 'The algorithm that produced the components or latent variables, which
  decides what their values and signs mean. Independent of `second_block`, since one
  block structure is reachable by several algorithms: a brain-behaviour covariance
  can be decomposed by PLS or by CCA. Open vocabulary: write the source''s own wording
  in the field when nothing here fits.'
in_subset:
- mvp
- model_extracted
from_schema: https://neurostuff.org/schemas/neuroimaging-paper-parse
rank: 1000
permissible_values:
  ica:
    text: ica
    description: Independent component analysis.
    in_subset:
    - mvp
    - model_extracted
  tensor_ica:
    text: tensor_ica
    description: ICA over a subject-by-time-by-space tensor, giving one shared spatial
      map with per-subject loadings.
    in_subset:
    - mvp
    - model_extracted
  joint_ica:
    text: joint_ica
    description: ICA on features concatenated across modalities, sharing one mixing
      matrix. Takes `other_imaging_modality` in `second_block`.
    in_subset:
    - mvp
    - model_extracted
  parallel_ica:
    text: parallel_ica
    description: ICA run on two blocks at once with their mixing profiles linked,
      so a component of one block is paired with a component of the other. Covers
      linked ICA.
    in_subset:
    - mvp
    - model_extracted
  pca:
    text: pca
    description: Principal component analysis, whose components are orthogonal rather
      than independent.
    in_subset:
    - mvp
    - model_extracted
  nmf:
    text: nmf
    description: Non-negative matrix factorization, whose components cannot take negative
      values.
    in_subset:
    - mvp
    - model_extracted
  dictionary_learning:
    text: dictionary_learning
    description: A sparse decomposition learning an overcomplete set of components.
    in_subset:
    - mvp
    - model_extracted
  pls:
    text: pls
    description: Partial least squares, where the source does not distinguish the
      correlation and regression variants.
    in_subset:
    - mvp
    - model_extracted
  pls_correlation:
    text: pls_correlation
    description: PLS correlation (PLS-C), decomposing the cross-block correlation
      matrix by SVD. The variant most fMRI work means by "PLS".
    in_subset:
    - mvp
    - model_extracted
  pls_regression:
    text: pls_regression
    description: PLS regression, which predicts one block from the other rather than
      decomposing their shared covariance symmetrically.
    in_subset:
    - mvp
    - model_extracted
  cca:
    text: cca
    description: Canonical correlation analysis, which maximizes correlation between
      blocks where PLS maximizes covariance.
    in_subset:
    - mvp
    - model_extracted
  mcca:
    text: mcca
    description: Multi-set CCA, over three or more blocks. Frequently chained into
      joint ICA as mCCA+jICA, which takes this value with `joint_ica` noted in `Analysis.model_representation_notes`
      when the chaining matters.
    in_subset:
    - mvp
    - model_extracted