mvpa2.clfs.warehouse.SplitRFE¶

class mvpa2.clfs.warehouse.SplitRFE(lrn, partitioner, fselector, errorfx=<function mean_mismatch_error>, fmeasure_postproc=None, fmeasure=None, nproc=1, **kwargs)¶

RFE with the nested cross-validation to estimate optimal number of features.

Given a learner (classifier) with a sensitivity analyzer and a partitioner, during training SplitRFE first performs a cross-validation with RFE to later estimate optimal number of features which should survive in RFE. Optimal number is chosen as the mid-point among all minimums of the average errors across splits. After deducing optimal number of features, SplitRFE applies regular RFE again on the full training dataset stopping at the estimated optimal number of features.

Notes

Available conditional attributes:

calling_time+: Time (in seconds) it took to call the node
errors+: History of errors
history+: Last step # when each feature was still present
nested_errors+: History of errors per each nested split
nested_nfeatures+: History of # of features left per each nested split
nfeatures+: History of # of features left
raw_results: Computed results before invoking postproc. Stored only if postproc is not None.
sensitivities: History of sensitivities (might consume too much memory
trained_dataset: The dataset it has been trained on
trained_nsamples+: Number of samples it has been trained on
trained_targets+: Set of unique targets (or any other space) it has been trained on (if present in the dataset trained on)
training_time+: Time (in seconds) it took to train the learner

(Conditional attributes enabled by default suffixed with +)

Examples

Resting on an example giving for the RFE here is an implementation using SplitRFE helper:

>>> # Lazy import
>>> from mvpa2.suite import *
>>> # design an RFE feature selection to be used with a classifier
>>> rfe = SplitRFE(
...           LinearCSVMC(),
...           OddEvenPartitioner(),
...           # take sensitivities per each split, L2 norm, abs, mean them
...           fmeasure_postproc=ChainMapper([
...               FxMapper('features', l2_normed),
...               FxMapper('samples', np.abs),
...               FxMapper('samples', np.mean)]),
...           # select 50% of the best on each step
...           fselector=FractionTailSelector(
...               0.50,
...               mode='select', tail='upper'),
...           # but we do want to update sensitivities on each step
...           update_sensitivity=True)
>>> clf = FeatureSelectionClassifier(
...           LinearCSVMC(),
...           # on features selected via RFE
...           rfe,
...           # custom description
...           descr='LinSVM+RFE(splits_avg)' )

But not only classifiers and their sensitivites could be used for RFE. It could be used even with univariate measures (e.g. OnewayAnova).

Attributes

`auto_train`	Whether the Learner performs automatic trainingwhen called untrained.
`bestdetector`
`descr`	Description of the object if any
`fmeasure`
`force_train`	Whether the Learner enforces training upon every call.
`fselector`
`is_trained`	Whether the Learner is currently trained.
`lrn`
`nfeatures_min`
`pass_attr`	Which attributes of the dataset or self.ca to pass into result dataset upon call
`pmeasure`
`postproc`	Node to perform post-processing of results
`slicearg`
`space`	Processing space name of this node
`splitter`
`stopping_criterion`
`train_pmeasure`
`update_sensitivity`

Methods

`__call__`(ds)
`forward`(data)	Map data from input to output space.
`forward1`(data)	Wrapper method to map single samples.
`generate`(ds)	Yield processing results.
`get_postproc`()	Returns the post-processing node or None.
`get_space`()	Query the processing space name of this node.
`reset`()
`reverse`(data)	Reverse-map data from output back into input space.
`reverse1`(data)
`set_postproc`(node)	Assigns a post-processing node
`set_space`(name)	Set the processing space name of this node.
`train`(ds)	The default implementation calls `_pretrain()`, `_train()`, and finally `_posttrain()`.
`untrain`()	Reverts changes in the state of this node caused by previous training

Parameters:

lrn : Learner

Learner with a sensitivity analyzer which will be used both for the sensitivity analysis and transfer error estimation

partitioner : Partitioner

Used to generate cross-validation partitions for cross-validation to deduce optimal number of features to maintain

fselector : Functor

Given a sensitivity map it has to return the ids of those features that should be kept.

errorfx : func, optional

Functor to use for estimation of cross-validation error

fmeasure_postproc : func, optional

Function to provide to the sensitivity analyzer as postproc. If no fmeasure is provided and classifier sensitivity is used, then maxofabs_sample() would be used for this postproc, unless other value is provided

enable_ca : None or list of str

Names of the conditional attributes which should be enabled in addition to the default ones

disable_ca : None or list of str

Names of the conditional attributes which should be disabled

update_sensitivity : bool

If False the sensitivity map is only computed once and reused for each iteration. Otherwise the sensitivities are recomputed at each selection step.

filler : optional

Value to fill empty entries upon reverse operation

auto_train : bool

Flag whether the learner will automatically train itself on the input dataset when called untrained.

force_train : bool

Flag whether the learner will enforce training on the input dataset upon every call.

space : str, optional

Name of the ‘processing space’. The actual meaning of this argument heavily depends on the sub-class implementation. In general, this is a trigger that tells the node to compute and store information about the input data that is “interesting” in the context of the corresponding processing in the output dataset.

pass_attr : str, list of str|tuple, optional

Additional attributes to pass on to an output dataset. Attributes can be taken from all three attribute collections of an input dataset (sa, fa, a – see Dataset.get_attr()), or from the collection of conditional attributes (ca) of a node instance. Corresponding collection name prefixes should be used to identify attributes, e.g. ‘ca.null_prob’ for the conditional attribute ‘null_prob’, or ‘fa.stats’ for the feature attribute stats. In addition to a plain attribute identifier it is possible to use a tuple to trigger more complex operations. The first tuple element is the attribute identifier, as described before. The second element is the name of the target attribute collection (sa, fa, or a). The third element is the axis number of a multidimensional array that shall be swapped with the current first axis. The fourth element is a new name that shall be used for an attribute in the output dataset. Example: (‘ca.null_prob’, ‘fa’, 1, ‘pvalues’) will take the conditional attribute ‘null_prob’ and store it as a feature attribute ‘pvalues’, while swapping the first and second axes. Simplified instructions can be given by leaving out consecutive tuple elements starting from the end.

postproc : Node instance, optional

Node to perform post-processing of results. This node is applied in __call__() to perform a final processing step on the to be result dataset. If None, nothing is done.

descr : str

Description of the instance

Attributes

`auto_train`	Whether the Learner performs automatic trainingwhen called untrained.
`bestdetector`
`descr`	Description of the object if any
`fmeasure`
`force_train`	Whether the Learner enforces training upon every call.
`fselector`
`is_trained`	Whether the Learner is currently trained.
`lrn`
`nfeatures_min`
`pass_attr`	Which attributes of the dataset or self.ca to pass into result dataset upon call
`pmeasure`
`postproc`	Node to perform post-processing of results
`slicearg`
`space`	Processing space name of this node
`splitter`
`stopping_criterion`
`train_pmeasure`
`update_sensitivity`

Methods

`__call__`(ds)
`forward`(data)	Map data from input to output space.
`forward1`(data)	Wrapper method to map single samples.
`generate`(ds)	Yield processing results.
`get_postproc`()	Returns the post-processing node or None.
`get_space`()	Query the processing space name of this node.
`reset`()
`reverse`(data)	Reverse-map data from output back into input space.
`reverse1`(data)
`set_postproc`(node)	Assigns a post-processing node
`set_space`(name)	Set the processing space name of this node.
`train`(ds)	The default implementation calls `_pretrain()`, `_train()`, and finally `_posttrain()`.
`untrain`()	Reverts changes in the state of this node caused by previous training

lrn¶