<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p>Is that Jet?!</p>
    <p><a class="moz-txt-link-freetext" href="https://www.youtube.com/watch?v=xAoljeRJ3lU">https://www.youtube.com/watch?v=xAoljeRJ3lU</a></p>
    <p>;)<br>
    </p>
    <div class="moz-cite-prefix">On 6/4/18 11:56 AM, Brown J.B. via
      scikit-learn wrote:<br>
    </div>
    <blockquote type="cite"
cite="mid:CAJe_vxBppv_1mbU6BTYiA0kJH+GYVucMq1-Kibwpb2uNt6q+Ew@mail.gmail.com">
      <div dir="ltr">
        <div>Hello community,</div>
        <div class="gmail_extra"><br>
          <div class="gmail_quote">
            <blockquote class="gmail_quote" style="margin:0px 0px 0px
              0.8ex;border-left:1px solid
              rgb(204,204,204);padding-left:1ex"><span class="gmail-">
                <blockquote class="gmail_quote" style="margin:0px 0px
                  0px 0.8ex;border-left:1px solid
                  rgb(204,204,204);padding-left:1ex">
                  I wonder if there's something similar for the binary
                  class case where,<br>
                  the prediction is a real value (activation) and from
                  this we can also<br>
                  derive<br>
                    - CMs for all prediction cutoff (or set of cutoffs?)<br>
                    - scores over all cutoffs (AUC, AP, ...)<br>
                </blockquote>
              </span>
              AUC and AP are by definition over all cut-offs. And CMs
              for all<br>
              cutoffs doesn't seem a good idea, because that'll be
              n_samples many<br>
              in the general case. If you want to specify a set of
              cutoffs, that would be pretty easy to do.<br>
              How do you find these cut-offs, though?<span
                class="gmail-"><br>
                <blockquote class="gmail_quote" style="margin:0px 0px
                  0px 0.8ex;border-left:1px solid
                  rgb(204,204,204);padding-left:1ex">
                  <br>
                  For me, in analyzing (binary class) performance,
                  reporting scores for<br>
                  a single cutoff is less useful than seeing how the
                  many scores (tpr,<br>
                  ppv, mcc, relative risk, chi^2, ...) vary at various
                  false positive<br>
                  rates, or prediction quantiles.<br>
                </blockquote>
              </span></blockquote>
            <div><br>
            </div>
            <div>In terms of finding cut-offs, one could use the idea of
              metric surfaces that I recently proposed</div>
            <div><a
                href="https://onlinelibrary.wiley.com/doi/abs/10.1002/minf.201700127"
                moz-do-not-send="true">https://onlinelibrary.wiley.com/doi/abs/10.1002/minf.201700127</a>
              <br>
            </div>
            <div>and then plot your per-threshold TPR/TNR pairs on the
              PPV/MCC/etc surfaces to determine what conditions you are
              willing to accept against the background of your
              prediction problem.</div>
            <div><br>
            </div>
            <div>I use these surfaces (a) to think about the prediction
              problem before any attempt at modeling is made, and (b) to
              deconstruct results such as "Accuracy=85%" into
              interpretations in the context of my field and the data
              being predicted.</div>
            <div><br>
            </div>
            <div>Hope this contributes a bit of food for thought.</div>
            <div>J.B.<br>
            </div>
          </div>
        </div>
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <pre class="moz-quote-pre" wrap="">_______________________________________________
scikit-learn mailing list
<a class="moz-txt-link-abbreviated" href="mailto:scikit-learn@python.org">scikit-learn@python.org</a>
<a class="moz-txt-link-freetext" href="https://mail.python.org/mailman/listinfo/scikit-learn">https://mail.python.org/mailman/listinfo/scikit-learn</a>
</pre>
    </blockquote>
  </body>
</html>