This block uses agglomerative information bottleneck, as proposed in Slonim et. al 2000. See also BLOCK_AIBDICT()
BK = BLOCK_AIB() Initializes the block with the default options.
BK = BLOCK_AIB(BK) Executes the block with options and inputs BK.
Required Inputs:
- db
A database partitioned into testing and training data.
- hist
Histograms to be compressed.
Options:
- bk.normalize_hists
Should histograms be normalized before they are accumulated for AIB? Default 1.
- bk.seg_ids
The segment ids to use for AIB. Default is all training examples.
- bk.rand_seed
Set the random seed. Default is [], which does not change the seed.
Fetchable attributes:
- Pcx
The joint class / dictionary probability matrix passed to AIB
tree:
The full AIB merge tree.