Update README.md

README.md

@@ -3120,4 +3120,42 @@ configs:
 
 Instance level results for assessors models trained on the [AFRLA - Instance Level Results](https://huggingface.co/datasets/DaniFrame/AFRLA-instance-level-results) dataset.
 
-At the moment of upload, results for XGBoost and linear regression models are available, with results from the former in 5 different seeds.
+At the moment of upload, results for XGBoost and linear regression models are available, with results from the former in 5 different seeds. Results are available for all 11 tasks described in the original dataset as well as for 6 different types of error (losses):
+
+<table style="width:50%; border-collapse: collapse;">
+  <tr>
+    <th> <center> Loss name </center> </th>
+    <th style="width:80%;"> <center> Description </center> </th>
+    <th style="width:75%"> <center> Formula </center> </th>
+  </tr>
+  <tr>
+    <td> <center> L<sub>1</sub> <sup>∓</sup> </center> </td>
+    <td> <center> Difference error with sign </center> </td>
+    <td> <center> y&#770; - y </center> </td>
+  </tr>
+  <tr>
+    <td> <center> L<sub>1</sub> <sup>+</sup> </center> </td>
+    <td> <center> Absolute error </center> </td>
+    <td> <center> |y&#770; - y|</center> </td>
+  </tr>
+  <tr>
+    <td> <center> L<sub>2</sub> <sup>∓</sup> </center> </td>
+    <td> <center> Squared error with sign </center> </td>
+    <td> <center> (y&#770; - y)<sup>2</sup> </center> </td>
+  </tr>
+  <tr>
+    <td> <center> L<sub>2</sub> <sup>+</sup> </center></td>
+    <td> <center> Squared error </center> </td>
+    <td> <center> (y&#770; - y)<sup>2</sup> · sgn(y&#770; - y) </center> </td>
+  </tr>
+  <tr>
+    <td> <center> L<sub>L</sub> <sup>∓</sup> </center> </td>
+    <td> <center> Logistic error with sign parametrised by a value β so that the mean absolute error is 0.5 </center></td>
+    <td> <center> 2 / (1+e<sup>-β(y&#770; - y)</sup>) - 1 </center></td>
+  </tr>
+  <tr>
+    <td> <center> L<sub>L</sub> <sup>+</sup> </center> </td>
+    <td> <center> Absolute logistic error parametrised by a value β so that the mean absolute error is 0.5 </center></td>
+    <td> <center> |2 / (1+e<sup>-β(y&#770; - y)</sup>) - 1| </center> </td>
+  </tr>
+</table>