(reproduce-tree-generation dt #!key global-splits)
Reconstruct the tree generation steps from the
data stored in the given decision tree and
displays the sequence of tree induction steps in the
proper order.
The output format looks like this:
0 [0.04531933508311461], split at (>= match-ratio 0.69047619047619)
1 [0.25773195876288657], split at (>= lit-ratio-ub 0.91287878787879)
2 [0.025802752293577983], split at (>= ir-score 276.5)
3 [0.040851553509781355], split at (>= match-ratio 0.41428571428571503)
4 [0.04716227018385292], split at (>= span-focus 0.5)
5 [0.18401937046004843], split at (>= focus-otype-matches 0.5)
Each line starts with the iteration index, which corresponds
to the current number of decision nodes (the first index is
0 since there is only a single leaf node at the beginning).
The number in square brackets is the probability of the YES
class in the leaf chosen for splitting.
The chosen split conditon is shown after split at,
e.g. (>= ir-score 276.5).
Parameters:
- dt
The given decision tree
The following DSSSL option key is supported:
- :global-splits
If :global-splits #t is specified, then the
first number in the generated lines is not the local
split count of inducing the given tree,
but rather the global split count in a parallel
learning situation
(co-induce-dt or co-induce-dt-bag)
in which the split was performed an in which the new
decision node was introduced.
In particular, the iteration indices shown at the
beginning of each lines are no longer subsequent numbers
in this case. When a gap occurs between shown global
iteration
indices, then another tree was extended by a new decision
node in the corresponding iteration.