ai-edge-quantizer-nightly 0.4.0.dev20251108__py3-none-any.whl → 0.4.0.dev20251110__py3-none-any.whl

ai_edge_quantizer/utils/validation_utils.py

@@ -32,7 +32,7 @@ def get_validation_func(
     a validation function
 
   Raises:
-    Value error if the function name is not supported
+    ValueError: if the function name is not supported
   """
   if func_name == "mse":
     return mean_squared_difference
@@ -42,6 +42,8 @@ def get_validation_func(
     return cosine_similarity
   elif func_name == "kl_divergence":
     return kl_divergence
+  elif func_name == "snr":
+    return signal_to_noise_ratio
   else:
     raise ValueError(f"Validation function {func_name} not supported")
 
@@ -62,7 +64,7 @@ def mean_squared_difference(
     a float value representing the MSD between data1 & 2
 
   Raises:
-    Value error if the two inputs don't have the same number of elements
+    ValueError: if the two inputs don't have the same number of elements
   """
   data1, data2 = _preprocess_same_size_arrays(data1, data2)
   # special handling for tensor of size 0
@@ -91,7 +93,7 @@ def median_diff_ratio(
     a float value representing the median diff ratio between data1 & 2
 
   Raises:
-    Value error if the two inputs don't have the same number of elements
+    ValueError: if the two inputs don't have the same number of elements
   """
   data1, data2 = _preprocess_same_size_arrays(data1, data2)
   # special handling for tensor of size 0
@@ -120,7 +122,7 @@ def cosine_similarity(
     a float value representing the cosine similarity between data1 & 2
 
   Raises:
-    Value error if the two inputs don't have the same number of elements
+    ValueError: if the two inputs don't have the same number of elements
   """
   data1, data2 = _preprocess_same_size_arrays(data1, data2)
   # special handling for tensor of size 0
@@ -152,15 +154,15 @@ def kl_divergence(
 
   Args:
     data1: input data to be used for comparison (distribution Q)
-    data2: input data to be used for comparison (distribution P),
-      data1 & 2 must be of the same shape
+    data2: input data to be used for comparison (distribution P), data1 & 2 must
+      be of the same shape
     epsilon: small value to avoid log(0) and division by zero.
 
   Returns:
     A float value representing the KL divergence between data1 & 2.
 
   Raises:
-    Value error if the two inputs don't have the same number of elements.
+    ValueError: if the two inputs don't have the same number of elements.
   """
   data1, data2 = _preprocess_same_size_arrays(data1, data2)
   # special handling for tensor of size 0
@@ -173,6 +175,40 @@ def kl_divergence(
   return float(np.sum(p * np.log((p + epsilon) / (q + epsilon))))
 
 
+def signal_to_noise_ratio(
+    noisy_signal: np._typing.ArrayLike,
+    signal: np._typing.ArrayLike,
+    epsilon: float = 1e-9,
+) -> float:
+  """Calculates the signal to noise ratio between noisy_signal & signal.
+
+  SNR = P_signal / P_noise, where signal is treated as the clean signal and
+  noisy_signal-signal is treated as the noise samples.
+  P_signal = mean(signal^2)
+  P_noise = mean((noisy_signal-signal)^2) = mse(noisy_signal, signal)
+
+  Args:
+    noisy_signal: Input data to be used for comparison (e.g. noisy signal).
+    signal: Input data to be used for comparison (e.g. clean signal),
+      noisy_signal & signal must be of the same shape.
+    epsilon: Small value to avoid division by zero.
+
+  Returns:
+    A float value representing the SNR between noisy_signal & signal.
+
+  Raises:
+    ValueError: If the two inputs don't have the same number of elements.
+  """
+  noisy_signal, signal = _preprocess_same_size_arrays(noisy_signal, signal)
+  if signal.size == 0:
+    return float(0)
+
+  mse = mean_squared_difference(noisy_signal, signal)
+  signal_power = float(np.square(signal).mean())
+  snr = signal_power / (mse + epsilon)
+  return snr
+
+
 def _preprocess_same_size_arrays(
     data1: np._typing.ArrayLike, data2: np._typing.ArrayLike
 ) -> Tuple[np.ndarray, np.ndarray]:
@@ -187,7 +223,7 @@ def _preprocess_same_size_arrays(
     a tuple of the preprocessed data1 & 2
 
   Raises:
-    Value error if the two inputs don't have the same number of elements
+    ValueError: if the two inputs don't have the same number of elements
   """
   data1 = np.array(data1, dtype=np.float32).flatten()
   data2 = np.array(data2, dtype=np.float32).flatten()

ai_edge_quantizer/utils/validation_utils_test.py

@@ -134,6 +134,34 @@ class ValidationUtilTest(googletest.TestCase):
     func = validation_utils.get_validation_func("kl_divergence")
     self.assertEqual(func, validation_utils.kl_divergence)
 
+  def test_signal_to_noise_ratio_0d(self):
+    data1 = []
+    data2 = []
+    result = validation_utils.signal_to_noise_ratio(data1, data2)
+    self.assertEqual(result, 0)
+
+  def test_signal_to_noise_ratio_identical(self):
+    data1 = [1, 2, 3]
+    data2 = [1, 2, 3]
+    result = validation_utils.signal_to_noise_ratio(data1, data2)
+    self.assertGreater(result, 1e8)  # mse=0, so snr should be large
+
+  def test_signal_to_noise_ratio_with_noise(self):
+    data1 = [2, 3, 4]
+    data2 = [1, 2, 3]
+    result = validation_utils.signal_to_noise_ratio(data1, data2)
+    self.assertAlmostEqual(result, 14 / 3, places=5)
+
+  def test_signal_to_noise_ratio_simple(self):
+    data1 = [1, 1]
+    data2 = [1, 0]
+    result = validation_utils.signal_to_noise_ratio(data1, data2)
+    self.assertAlmostEqual(result, 1.0, places=5)
+
+  def test_get_validation_func_snr(self):
+    func = validation_utils.get_validation_func("snr")
+    self.assertEqual(func, validation_utils.signal_to_noise_ratio)
+
 
 if __name__ == "__main__":
   googletest.main()

{ai_edge_quantizer_nightly-0.4.0.dev20251108.dist-info → ai_edge_quantizer_nightly-0.4.0.dev20251110.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ai-edge-quantizer-nightly
-Version: 0.4.0.dev20251108
+Version: 0.4.0.dev20251110
 Summary: A quantizer for advanced developers to quantize converted AI Edge models.
 Home-page: https://github.com/google-ai-edge/ai-edge-quantizer
 Keywords: On-Device ML,AI,Google,TFLite,Quantization,LLMs,GenAI

{ai_edge_quantizer_nightly-0.4.0.dev20251108.dist-info → ai_edge_quantizer_nightly-0.4.0.dev20251110.dist-info}/RECORD

@@ -72,10 +72,10 @@ ai_edge_quantizer/utils/tfl_flatbuffer_utils.py,sha256=42OWzQsRTXq3XQYmoxlz177_d
 ai_edge_quantizer/utils/tfl_flatbuffer_utils_test.py,sha256=K1SbK8q92qYVtiVj0I0GtugsPTkpIpEKv9zakvFV_Sc,8555
 ai_edge_quantizer/utils/tfl_interpreter_utils.py,sha256=EoVjI_hplX_Rml3hfRsGmQOihexmizeJqt4SQcET9aA,14925
 ai_edge_quantizer/utils/tfl_interpreter_utils_test.py,sha256=6fjkM-rycZ95L4yfvlr0TN6RlrhfPzxNUYrZaYO_F0A,12013
-ai_edge_quantizer/utils/validation_utils.py,sha256=QTYyQ_HDVrFTGPIsrA240Lv8tUw1fwWp2fu9kTVISkE,6224
-ai_edge_quantizer/utils/validation_utils_test.py,sha256=lO51rGskhzpXePRdZMU87u_YO35_sDp9_eQ85CmupL4,4600
-ai_edge_quantizer_nightly-0.4.0.dev20251108.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-ai_edge_quantizer_nightly-0.4.0.dev20251108.dist-info/METADATA,sha256=yUTFddw1QJHge_e5fD2irNEXLnRdCw1SSBWtB7pBPcY,1707
-ai_edge_quantizer_nightly-0.4.0.dev20251108.dist-info/WHEEL,sha256=SmOxYU7pzNKBqASvQJ7DjX3XGUF92lrGhMb3R6_iiqI,91
-ai_edge_quantizer_nightly-0.4.0.dev20251108.dist-info/top_level.txt,sha256=8QTfPnFXNVUhScFLaa-NWZMFWMn72M50DVPubpwWB1g,18
-ai_edge_quantizer_nightly-0.4.0.dev20251108.dist-info/RECORD,,
+ai_edge_quantizer/utils/validation_utils.py,sha256=Mr0D6X-pTDLODFAnCX3IlqdV1OL02tlq0ZjHbqx8nzg,7439
+ai_edge_quantizer/utils/validation_utils_test.py,sha256=T8K5mCWeMcihND2KS_dHvCJUU9lEdG2sD95EgPkaX3w,5584
+ai_edge_quantizer_nightly-0.4.0.dev20251110.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+ai_edge_quantizer_nightly-0.4.0.dev20251110.dist-info/METADATA,sha256=se3xo08IXOZLI8qd_7g_aH_o32b84BLW223gw1nQYS4,1707
+ai_edge_quantizer_nightly-0.4.0.dev20251110.dist-info/WHEEL,sha256=SmOxYU7pzNKBqASvQJ7DjX3XGUF92lrGhMb3R6_iiqI,91
+ai_edge_quantizer_nightly-0.4.0.dev20251110.dist-info/top_level.txt,sha256=8QTfPnFXNVUhScFLaa-NWZMFWMn72M50DVPubpwWB1g,18
+ai_edge_quantizer_nightly-0.4.0.dev20251110.dist-info/RECORD,,