[ https://issues.apache.org/jira/browse/SYSTEMML-1678?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16067098#comment-16067098 ]
Mike Dusenberry commented on SYSTEMML-1678: ------------------------------------------- cc [~niketanpansare] We could use this in Caffe2DML to compute the top predictions in the normal 1D case. > Add new 1D top_k utility function > --------------------------------- > > Key: SYSTEMML-1678 > URL: https://issues.apache.org/jira/browse/SYSTEMML-1678 > Project: SystemML > Issue Type: Sub-task > Reporter: Mike Dusenberry > Assignee: Fei Hu > Fix For: SystemML 1.0 > > > We should add new {{top_k}} and {{top_k2d}} utility functions (in > {{nn/util.dml}}) that accept a matrix {{X}} and return matrices {{values}} > and {{indices}} with the top {{k}} values (i.e. probabilities) and associated > indices (i.e. classes) along a certain dimension. This will be modeled after > the [{{top_k}} function in TensorFlow | > https://www.tensorflow.org/api_docs/python/tf/nn/top_k] For the 1D case, > {{top_k}} will operate on the columns dimension. A typical use case is that > in which {{X}} is the output of a {{softmax}} layer (so each row contains a > set of normalized class probabilities), and {{values}} and {{indices}} will > contain rows with the top {{k}} probabilities and class indices as described > above. For the 2D case, {{top_k}} will operate on the channels dimension. A > typical use case here is that in which {{X}} is the output of a {{softmax2d}} > layer (so each channel contains a set of normalized class probabilities), and > {{values}} and {{indices}} will contain the top {{k}} probabilities and > indices along the channel axis. This scenario would be common in an image > segmentation problem, in which every pixel of the output image will have a > set of class probabilities along the channel axis. > Having these {{top-k}} functions will allow us to extract either predict a > single class for each item, or the top {{k}} classes, and therefore may be > more useful that a {{predict_class}} function. > Although we will use {{values}} and {{indices}} as the names of the returned > matrices within the functions, in practice, one is likely to name the results > {{probs}} and {{classes}} in the calling environment. -- This message was sent by Atlassian JIRA (v6.4.14#64029)