<div dir="ltr"><br clear="all"><div><span style="font-family:arial,sans-serif;font-size:12.727272033691406px">Hello all,</span><div style="font-family:arial,sans-serif;font-size:12.727272033691406px"><br></div><div style="font-family:arial,sans-serif;font-size:12.727272033691406px">
on line 1279 of mvpa2/clfs/meta.py it reads:</div><div style="font-family:arial,sans-serif;font-size:12.727272033691406px"><div>class MappedClassifier(ProxyClassifier):</div><div>    """`ProxyClassifier` which uses some mapper prior training/testing.                                                                                      </div>
<div>                                                                                                                                                             </div><div>    `MaskMapper` can be used just a subset of features to                                                                                                    </div>
<div>    train/classify.                                                                                                                                          </div><div>    Having such classifier we can easily create a set of classifiers                                                                                         </div>
<div>    for BoostedClassifier, where each classifier operates on some set                                                                                        </div><div>    of features, e.g. set of best spheres from SearchLight, set of                                                                                           </div>
<div>    ROIs selected elsewhere. It would be different from simply                                                                                               </div><div>    applying whole mask over the dataset, since here initial decision                                                                                        </div>
<div>    is made by each classifier and then later on they vote for the                                                                                           </div><div>    final decision across the set of classifiers.                                                                                                            </div>
<div>    """</div><div><br></div><div>Has anyone accomplished this?</div><div>In other words:</div><div>1) Data: 50 features, each with 10000 observations</div><div>2) Train several clfs, each with its own optimized feature selection method; because each classifier will prefer a distinct set of features, overall feature coverage after feature selection will be better if many classifiers are combined.</div>
<div>3) Combine function</div><div>4) According to the literature, the combined classifiers (even if many of them are weak) should lead to a dramatic improvement in classification accuracy.</div><div>5) Right now I can get ~65% accuracy on a three-fold cross validation with SMLR(), but am hoping to get near 80% with a combined approach.</div>
<div>6) It does not have to be the one stated in meta.py, but this seems like a good lead.</div><div><br></div><div>Any help is much appreciated!</div><div style>Thanks!</div></div></div><b><span style="font-family:arial,sans-serif;font-size:13px;font-weight:normal;background-color:rgb(255,255,255)"><b><span style="font-family:arial,sans-serif;font-size:13px;font-weight:normal;background-color:rgb(255,255,255)"><font color="#999999"><font color="#888888"><br>
</font></font></span></b></span></b>
</div>