0.17 patches (#325)

* fixes #323 Default indexer distance is now cosine in Sim Model.

Calling create_index method now defaults to cosine distance.

Additionally, auto distance defaults to cosine if no distance is passed to compile.

* fixes #322 remove all calls to tf.convert_to_tensor in SimModel.

* Update gitignore to exclude models and datasets from the example notebooks.

* Update multi-modal notebook to remove the call to compile.

* Patch bump

.gitignore

@@ -11,6 +11,8 @@ release.sh
 benchmark/supervised/datasets/
 benchmark/supervised/models/
 datasets/
+multi_modal_datasets/
+*.h5
 
 # Byte-compiled / optimized / DLL files
 __pycache__/

tensorflow_similarity/__init__.py

@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-__version__ = "0.17.0"
+__version__ = "0.17.1"
 
 
 from . import algebra  # noqa

tensorflow_similarity/models/similarity_model.py

@@ -198,17 +198,17 @@ def compile(
         """
         # Fetching the distance used from the first loss if auto
         if distance == "auto":
-            if isinstance(loss, list):
-                metric_loss = loss[0]
+            if loss is None:
+                distance = distance_canonicalizer("cosine")
             else:
-                metric_loss = loss
+                metric_loss = loss[0] if isinstance(loss, list) else loss
 
-            try:
-                distance = metric_loss.distance
-            except AttributeError:
-                msg = "distance='auto' only works if the first loss is a " "metric loss"
+                try:
+                    distance = metric_loss.distance
+                except AttributeError:
+                    msg = "distance='auto' only works if the first loss is a " "metric loss"
+                    raise ValueError(msg)
 
-                raise ValueError(msg)
             print(f"Distance metric automatically set to {distance} use the " "distance arg to override.")
         else:
             distance = distance_canonicalizer(distance)
@@ -265,7 +265,7 @@ def _index(self, index: Indexer) -> None:
 
     def create_index(
         self,
-        distance: Distance | str = "auto",
+        distance: Distance | str = "cosine",
         search: Search | str = "nmslib",
         kv_store: Store | str = "memory",
         evaluator: Evaluator | str = "memory",
@@ -362,8 +362,6 @@ def index(
         if verbose:
             print(f"[Indexing {tf.shape(x)[0]} points]")
             print("|-Computing embeddings")
-        with tf.device("/cpu:0"):
-            x = tf.convert_to_tensor(np.array(x))
 
         predictions = self.predict(x)
 
@@ -430,8 +428,6 @@ def lookup(self, x: Tensor, k: int = 5, verbose: int = 1) -> list[list[Lookup]]:
             list of list of k nearest neighboors:
             list[list[Lookup]]
         """
-        with tf.device("/cpu:0"):
-            x = tf.convert_to_tensor(np.array(x))
         predictions = self.predict(x)
 
         return self._index.batch_lookup(predictions=predictions, k=k, verbose=verbose)
@@ -516,8 +512,6 @@ def calibrate(
             thresholds_targets = {}
 
         # predict
-        with tf.device("/cpu:0"):
-            x = tf.convert_to_tensor(np.array(x))
         predictions = self.predict(x)
 
         # calibrate
@@ -582,8 +576,6 @@ def match(
             raise ValueError("Uncalibrated model: run model.calibration()")
 
         # get predictions
-        with tf.device("/cpu:0"):
-            x = tf.convert_to_tensor(np.array(x))
         predictions = self.predict(x)
 
         # matching
@@ -634,8 +626,6 @@ def evaluate_retrieval(
         # get embeddings
         if verbose:
             print("|-Computing embeddings")
-        with tf.device("/cpu:0"):
-            x = tf.convert_to_tensor(np.array(x))
         predictions = self.predict(x)
 
         if verbose:
@@ -713,8 +703,6 @@ def evaluate_classification(
         # get embeddings
         if verbose:
             print("|-Computing embeddings")
-        with tf.device("/cpu:0"):
-            x = tf.convert_to_tensor(np.array(x))
         predictions = self.predict(x)
 
         results: defaultdict[str, dict[str, str | np.ndarray]] = defaultdict(dict)