This is an automated email from the ASF dual-hosted git repository.
mboehm7 pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/systemds.git
commit c7f53c0b22494a290bfd8f251040edc566e85af6
Author: Matthias Boehm <mboe...@gmail.com>
AuthorDate: Sat Apr 20 18:00:35 2024 +0200
[SYSTEMDS-3538] Add missing f1Score builtin function
We simply compute the F1 score based on the confusion matrix output,
and only support binary labels. In the future, we could adopt an
one-against-the-rest approach for multi-class settings.
---
scripts/builtin/confusionMatrix.dml | 2 +-
.../builtin/{confusionMatrix.dml => f1Score.dml} | 39 ++++++----------------
.../java/org/apache/sysds/common/Builtins.java | 1 +
3 files changed, 12 insertions(+), 30 deletions(-)
diff --git a/scripts/builtin/confusionMatrix.dml
b/scripts/builtin/confusionMatrix.dml
index 14d176706c..3ac70fb3f8 100644
--- a/scripts/builtin/confusionMatrix.dml
+++ b/scripts/builtin/confusionMatrix.dml
@@ -34,7 +34,7 @@
# INPUT:
#
------------------------------------------------------------------------------
# P vector of predictions (1-based, recoded)
-# Y vector of actual labels (1-based, recorded)
+# Y vector of actual labels (1-based, recoded)
#
------------------------------------------------------------------------------
#
# OUTPUT:
diff --git a/scripts/builtin/confusionMatrix.dml b/scripts/builtin/f1Score.dml
similarity index 50%
copy from scripts/builtin/confusionMatrix.dml
copy to scripts/builtin/f1Score.dml
index 14d176706c..0407998ac1 100644
--- a/scripts/builtin/confusionMatrix.dml
+++ b/scripts/builtin/f1Score.dml
@@ -19,44 +19,25 @@
#
#-------------------------------------------------------------
-# Computes the confusion matrix for input vectors of predictions
-# and actual labels. We return both the counts and relative frequency
-# (normalized by sum of true labels)
-#
-# .. code-block::
-#
-# True Labels
-# 1 2
-# 1 TP | FP
-# Predictions ----+----
-# 2 FN | TN
+# Computes the F1 score as the harmonic mean of precision and recall.
+# F1 = 2TP / (2TP + FP + FN)
#
# INPUT:
#
------------------------------------------------------------------------------
# P vector of predictions (1-based, recoded)
-# Y vector of actual labels (1-based, recorded)
+# Y vector of actual labels (1-based, recoded)
#
------------------------------------------------------------------------------
#
# OUTPUT:
#
------------------------------------------------------------------------------
-# confusionSum the confusion matrix as absolute counts
-# confusionAvg the confusion matrix as relative frequencies
+# score the F1 score
#
------------------------------------------------------------------------------
-m_confusionMatrix = function(Matrix[Double] P, Matrix[Double] Y)
- return(Matrix[Double] confusionSum, Matrix[Double] confusionAvg)
+m_f1Score = function(Matrix[Double] P, Matrix[Double] Y)
+ return(Double score)
{
- dim = max(max(Y), max(P)) #ensure known dim
-
- if(ncol(P) > 1 | ncol(Y) > 1)
- stop("confusionMatrix: Invalid input number of cols should be 1 in both P
["+ncol(P)+"] and Y ["+ncol(Y)+"]")
- if(nrow(P) != nrow(Y))
- stop("confusionMatrix: The number of rows have to be equal in both P
["+nrow(P)+"] and Y ["+nrow(Y)+"]")
- if(min(P) < 1 | min(Y) < 1)
- stop("confusionMatrix: All Values in P and Y should be abore or equal to
1, min(P):" + min(P) + " min(Y):" + min(Y) )
-
- confusionSum = table(P, Y, dim, dim)
- # max to avoid division by 0, in case a colum contain no entries.
- confusionAvg = confusionSum / max(1,colSums(confusionSum))
+ [cS, cA] = confusionMatrix(P, Y);
+ if(nrow(cS)>2 | ncol(cS)>2)
+ stop("f1Score: currently only supported for binary class labels.");
+ score = as.scalar(2*cS[1,1] / (2*cS[1,1] + cS[2,1] + cS[1,2]));
}
-
diff --git a/src/main/java/org/apache/sysds/common/Builtins.java
b/src/main/java/org/apache/sysds/common/Builtins.java
index b510698cd4..61faccc0bc 100644
--- a/src/main/java/org/apache/sysds/common/Builtins.java
+++ b/src/main/java/org/apache/sysds/common/Builtins.java
@@ -130,6 +130,7 @@ public enum Builtins {
EXP("exp", false),
EVAL("eval", false),
EVALLIST("evalList", false),
+ F1SCORE("f1Score", true),
FIT_PIPELINE("fit_pipeline", true),
FIX_INVALID_LENGTHS("fixInvalidLengths", true),
FIX_INVALID_LENGTHS_APPLY("fixInvalidLengthsApply", true),
