From 0540237d794ab2071764dc74e4d3bb52f5bf44be Mon Sep 17 00:00:00 2001
From: Gustaf Rydholm <gustaf.rydholm@gmail.com>
Date: Tue, 27 Sep 2022 00:10:26 +0200
Subject: Update metrics

---
 text_recognizer/models/metrics/cer.py | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
 create mode 100644 text_recognizer/models/metrics/cer.py

(limited to 'text_recognizer/models/metrics/cer.py')

diff --git a/text_recognizer/models/metrics/cer.py b/text_recognizer/models/metrics/cer.py
new file mode 100644
index 0000000..238ecc3
--- /dev/null
+++ b/text_recognizer/models/metrics/cer.py
@@ -0,0 +1,23 @@
+"""Character Error Rate (CER)."""
+from typing import Sequence
+
+import torch
+import torchmetrics
+
+
+class CharacterErrorRate(torchmetrics.CharErrorRate):
+    """Character error rate metric, allowing for tokens to be ignored."""
+
+    def __init__(self, ignore_tokens: Sequence[int], *args):
+        super().__init__(*args)
+        self.ignore_tokens = set(ignore_tokens)
+
+    def update(self, preds: torch.Tensor, targets: torch.Tensor) -> None:
+        preds_l = [
+            [t for t in pred if t not in self.ignore_tokens] for pred in preds.tolist()
+        ]
+        targets_l = [
+            [t for t in target if t not in self.ignore_tokens]
+            for target in targets.tolist()
+        ]
+        super().update(preds_l, targets_l)
-- 
cgit v1.2.3-70-g09d2