|132178||Total predicted (of 286241 candidates)|
Notes: The classifier only generates a prediction if its confidence exceeds 0.9.
Good candidates are single pulses that could not be matched to a known pulsar. They are the events most likely to lead to a new discovery; they should be reviewed first.
Of next importance to review are the candidates that the classifier could not confidently classify. (We hope to have a link here soon to those candidates as well.)
This plot shows the prediction performance for the V-FASTR classifier. For each month, a classifier was trained on all labeled data collected prior to that month. This classifier was then used to generate predictions for all labeled data collected in that month.
The blue line shows the accuracy of these predictions, and the green line shows the percentage of human effort that would be saved by using the classifier to filter out artifacts and known pulsars, reserving the remaining good candidates and unclassified candidates for human review.
This is a visualization of the sky pointing locations for candidates classified by the classifier as Artifact (left) or Pulsar (right). The size of each blue circle is proprtional to the number of candidates classified at that location.