Model tests use smaller learning rate to catch performance regression.

james-choncholas · Jan 20, 2025 · 6f6dea8 · 6f6dea8
1 parent 368f3b1
commit 6f6dea8
Show file tree

Hide file tree

Showing 7 changed files with 7 additions and 10 deletions.
diff --git a/tf_shell_ml/test/dpsgd_conv_model_local_test.py b/tf_shell_ml/test/dpsgd_conv_model_local_test.py
@@ -112,9 +112,6 @@ def _test_model(self, disable_encryption, disable_masking, disable_noise, cache)
             metrics=[tf.keras.metrics.CategoricalAccuracy()],
         )
 
-        m.build([None, 28 - (2 * clip_by), 28 - (2 * clip_by), 1])
-        m.summary()
-
         history = m.fit(
             features_dataset,
             labels_dataset,

diff --git a/tf_shell_ml/test/dpsgd_model_distrib_test.py b/tf_shell_ml/test/dpsgd_model_distrib_test.py
@@ -108,7 +108,7 @@ def test_model(self):
                 lambda read_from_cache: tf_shell.create_autocontext64(
                     log2_cleartext_sz=25,
                     scaling_factor=1,
-                    noise_offset_log2=2,
+                    noise_offset_log2=0,
                     read_from_cache=read_from_cache,
                     cache_path=cache,
                 ),
@@ -119,7 +119,7 @@ def test_model(self):
 
             m.compile(
                 loss=tf.keras.losses.CategoricalCrossentropy(),
-                optimizer=tf.keras.optimizers.Adam(0.1),
+                optimizer=tf.keras.optimizers.Adam(0.01, beta_1=0.8),
                 metrics=[tf.keras.metrics.CategoricalAccuracy()],
             )
 

diff --git a/tf_shell_ml/test/dpsgd_model_float64_test.py b/tf_shell_ml/test/dpsgd_model_float64_test.py
@@ -79,7 +79,7 @@ def _test_model(self, disable_encryption, disable_masking, disable_noise, cache)
 
         m.compile(
             loss=tf.keras.losses.CategoricalCrossentropy(),
-            optimizer=tf.keras.optimizers.Adam(0.1),
+            optimizer=tf.keras.optimizers.Adam(0.01, beta_1=0.8),
             metrics=[tf.keras.metrics.CategoricalAccuracy()],
         )
 

diff --git a/tf_shell_ml/test/dpsgd_model_local_test.py b/tf_shell_ml/test/dpsgd_model_local_test.py
@@ -79,7 +79,7 @@ def _test_model(self, disable_encryption, disable_masking, disable_noise, cache)
 
         m.compile(
             loss=tf.keras.losses.CategoricalCrossentropy(),
-            optimizer=tf.keras.optimizers.Adam(0.1),
+            optimizer=tf.keras.optimizers.Adam(0.01, beta_1=0.8),
             metrics=[tf.keras.metrics.CategoricalAccuracy()],
         )
 

diff --git a/tf_shell_ml/test/postscale_model_distrib_test.py b/tf_shell_ml/test/postscale_model_distrib_test.py
@@ -115,7 +115,7 @@ def test_model(self):
 
             m.compile(
                 loss=tf.keras.losses.CategoricalCrossentropy(),
-                optimizer=tf.keras.optimizers.Adam(0.1),
+                optimizer=tf.keras.optimizers.Adam(0.01, beta_1=0.8),
                 metrics=[tf.keras.metrics.CategoricalAccuracy()],
             )
 

diff --git a/tf_shell_ml/test/postscale_model_float64_test.py b/tf_shell_ml/test/postscale_model_float64_test.py
@@ -71,7 +71,7 @@ def _test_model(self, disable_encryption, disable_masking, disable_noise, cache)
 
         m.compile(
             loss=tf.keras.losses.CategoricalCrossentropy(),
-            optimizer=tf.keras.optimizers.Adam(0.1),
+            optimizer=tf.keras.optimizers.Adam(0.01, beta_1=0.8),
             metrics=[tf.keras.metrics.CategoricalAccuracy()],
         )
 

diff --git a/tf_shell_ml/test/postscale_model_local_test.py b/tf_shell_ml/test/postscale_model_local_test.py
@@ -71,7 +71,7 @@ def _test_model(self, disable_encryption, disable_masking, disable_noise, cache)
 
         m.compile(
             loss=tf.keras.losses.CategoricalCrossentropy(),
-            optimizer=tf.keras.optimizers.Adam(0.1),
+            optimizer=tf.keras.optimizers.Adam(0.01, beta_1=0.8),
             metrics=[tf.keras.metrics.CategoricalAccuracy()],
         )