Fix a few typos in comments (#21525)

Author: bzamecnik

keras/src/backend/jax/numpy.py

@@ -64,7 +64,7 @@ def kaiser(x, beta):
 
 
 def bincount(x, weights=None, minlength=0, sparse=False):
-    # Note: bincount is never tracable / jittable because the output shape
+    # Note: bincount is never traceable / jittable because the output shape
     # depends on the values in x.
     if sparse or isinstance(x, jax_sparse.BCOO):
         if isinstance(x, jax_sparse.BCOO):

keras/src/layers/attention/grouped_query_attention_test.py

@@ -293,7 +293,7 @@ def test_masking(self, use_causal_mask):
     )
     def test_correctness(self, flash_attention):
         if flash_attention:
-            # Let the backend decide whether to use flase attention
+            # Let the backend decide whether to use flash attention
             enable_flash_attention()
         dtype = "float16"  # Flash attention only accepts float16/bfloat16
         head_dim = 8  # key_dim % 8 == 0 to enable flash attention

keras/src/layers/attention/multi_head_attention_test.py

@@ -395,7 +395,7 @@ def test_no_warning_with_keras_mask(self):
     )
     def test_correctness(self, flash_attention):
         if flash_attention:
-            # Let the backend decide whether to use flase attention
+            # Let the backend decide whether to use flash attention
             enable_flash_attention()
         dtype = "float16"  # Flash attention only accepts float16/bfloat16
 

keras/src/layers/preprocessing/hashing.py

@@ -214,7 +214,7 @@ def call(self, inputs):
 
         inputs = tf_utils.ensure_tensor(inputs)
         if self.output_mode == "one_hot" and inputs.shape[-1] == 1:
-            # One hot only unpranks if the final dimension is not 1.
+            # One hot only upranks if the final dimension is not 1.
             inputs = tf_backend.numpy.squeeze(inputs, axis=-1)
         if isinstance(inputs, tf.SparseTensor):
             indices = tf.SparseTensor(

keras/src/metrics/confusion_metrics.py

@@ -1494,16 +1494,16 @@ def result(self):
             # 1) Both measures diverge when there are no negative values;
             # 2) Both measures diverge when there are no true positives;
             # 3) Recall gain becomes negative when the recall is lower than the
-            #    label average (i.e. when more negative exampless are
+            #    label average (i.e. when more negative examples are
             #    classified positive than real positives).
             #
             # We ignore case 1 as it is easily understood that metrics would be
             # badly defined then. For case 2 we set recall_gain to 0 and
             # precision_gain to 1. For case 3 we set recall_gain to 0. These
-            # fixes will result in an overstimation of the AUCfor estimators
+            # fixes will result in an overestimation of the AUC for estimators
             # that are anti-correlated with the label (at some threshold).
 
-            # The scaling factor $\frac{P}{N}$ that is used to for mboth gain
+            # The scaling factor $\frac{P}{N}$ that is used to for both gain
             # values.
             scaling_factor = ops.divide_no_nan(
                 ops.add(self.true_positives, self.false_negatives),

keras/src/metrics/confusion_metrics_test.py

@@ -1407,7 +1407,7 @@ def test_weighted_prgain_majoring(self):
         )
 
         # tp = [7, 4, 0], fp = [3, 0, 0], fn = [0, 3, 7], tn = [0, 3, 3]
-        # scaling_facor (P/N) = 7/3
+        # scaling_factor (P/N) = 7/3
         # recall_gain = 1 - 7/3 [0/7, 3/4, 7/0] = [1, -3/4, -inf] -> [1, 0, 0]
         # precision_gain = 1 - 7/3 [3/7, 0/4, 0/0] = [0, 1, NaN] -> [0, 1, 1]
         # heights = [max(0, 1), max(1, 1)] = [1, 1]
@@ -1426,7 +1426,7 @@ def test_weighted_prgain_minoring(self):
         )
 
         # tp = [7, 4, 0], fp = [3, 0, 0], fn = [0, 3, 7], tn = [0, 3, 3]
-        # scaling_facor (P/N) = 7/3
+        # scaling_factor (P/N) = 7/3
         # recall_gain = 1 - 7/3 [0/7, 3/4, 7/0] = [1, -3/4, -inf] -> [1, 0, 0]
         # precision_gain = 1 - 7/3 [3/7, 0/4, 0/0] = [0, 1, NaN] -> [0, 1, 1]
         # heights = [min(0, 1), min(1, 1)] = [0, 1]
@@ -1443,7 +1443,7 @@ def test_weighted_prgain_interpolation(self):
         )
 
         # tp = [7, 4, 0], fp = [3, 0, 0], fn = [0, 3, 7], tn = [0, 3, 3]
-        # scaling_facor (P/N) = 7/3
+        # scaling_factor (P/N) = 7/3
         # recall_gain = 1 - 7/3 [0/7, 3/4, 7/0] = [1, -3/4, -inf] -> [1, 0, 0]
         # precision_gain = 1 - 7/3 [3/7, 0/4, 0/0] = [0, 1, NaN] -> [0, 1, 1]
         # heights = [(0+1)/2, (1+1)/2] = [0.5, 1]
@@ -1461,7 +1461,7 @@ def test_prgain_interpolation(self):
         result = auc_obj(y_true, y_pred)
 
         # tp = [5, 3, 0], fp = [5, 1, 0], fn = [0, 2, 5], tn = [0, 4, 4]
-        # scaling_facor (P/N) = 5/5 = 1
+        # scaling_factor (P/N) = 5/5 = 1
         # recall_gain = 1 - [0/5, 2/3, 5/0] = [1, 1/3, -inf] -> [1, 1/3, 0]
         # precision_gain = 1 - [5/5, 1/3, 0/0] = [1, 1/3, NaN] -> [0, 2/3, 1]
         # heights = [(0+2/3)/2, (2/3+1)/2] = [0.333333, 0.833333]

keras/src/metrics/metrics_utils.py

@@ -318,7 +318,7 @@ def is_evenly_distributed_thresholds(thresholds):
     """Check if the thresholds list is evenly distributed.
 
     We could leverage evenly distributed thresholds to use less memory when
-    calculate metrcis like AUC where each individual threshold need to be
+    calculate metrics like AUC where each individual threshold need to be
     evaluated.
 
     Args:

keras/src/models/functional.py

@@ -298,7 +298,7 @@ def _standardize_inputs(self, inputs):
             self._inputs_struct, dict
         ):
             # This is to avoid warning
-            # when we have reconciable dict/list structs
+            # when we have reconcilable dict/list structs
             if hasattr(self._inputs_struct, "__len__") and all(
                 isinstance(i, backend.KerasTensor) for i in self._inputs_struct
             ):

keras/src/ops/math_test.py

@@ -857,7 +857,7 @@ def test_istft(
         )
         if backend.backend() in ("numpy", "jax", "torch"):
             # these backends have different implementation for the boundary of
-            # the output, so we need to truncate 5% befroe assertAllClose
+            # the output, so we need to truncate 5% before assertAllClose
             truncated_len = int(output.shape[-1] * 0.05)
             output = output[..., truncated_len:-truncated_len]
             ref = ref[..., truncated_len:-truncated_len]
@@ -889,7 +889,7 @@ def test_istft(
         )
         if backend.backend() in ("numpy", "jax", "torch"):
             # these backends have different implementation for the boundary of
-            # the output, so we need to truncate 5% befroe assertAllClose
+            # the output, so we need to truncate 5% before assertAllClose
             truncated_len = int(output.shape[-1] * 0.05)
             output = output[..., truncated_len:-truncated_len]
             ref = ref[..., truncated_len:-truncated_len]

keras/src/testing/test_case.py

@@ -525,7 +525,7 @@ def __call__(self, y_true, y_pred, sample_weight=None):
                 )
 
                 # Ensure that the subclass layer doesn't mark itself as built
-                # when `build` is overriden.
+                # when `build` is overridden.
 
                 class ModifiedBuildLayer(layer_cls):
                     def build(self, *args, **kwargs):