fixes #500

CHANGELOG.md

@@ -6,13 +6,13 @@ Most recent releases are shown at the top. Each release shows:
 - **Changed**: Additional parameters, changes to inputs or outputs, etc
 - **Fixed**: Bug fixes that don't change documented behaviour
 
-## 0.37.2 (TBD)
+## 0.37.2 (2023-06-14)
 
 ### new:
 - N/A
 
 ### changed
-- N/A
+- `text.models`, `vision.models`, and `tabular.models` now all automatically set metrics to use `binary_accuracy` for multilabel problems
 
 ### fixed:
 - fix `validate` to support multilabel classification problems (#498)

ktrain/tabular/models.py

@@ -23,7 +23,7 @@ def _tabular_model(
     train_data,
     multilabel=None,
     is_regression=False,
-    metrics=["accuracy"],
+    metrics=None,
     hidden_layers=[1000, 500],
     hidden_dropouts=[0.0, 0.5],
     bn=False,
@@ -40,7 +40,8 @@ def _tabular_model(
                             If false, binary/multiclass model will be returned.
                             If None, multilabel will be inferred from data.
         is_regression(bool): If True, will build a regression model, else classification model.
-        metrics(list): list of metrics to use
+        metrics(list): List of metrics to use.  If None: 'accuracy' is used for binar/multiclass,
+                       'binary_accuracy' is used for multilabel classification, and 'mae' is used for regressio
         hidden_layers(list): number of units in each hidden layer of NN
         hidden_dropouts(list): Dropout values after each hidden layer of NN
         bn(bool): If True, BatchNormalization will be used before each fully-connected layer in NN
@@ -73,8 +74,12 @@ def _tabular_model(
         loss_func = "mse"
         activation = "linear"
     else:  # classification
-        if metrics is None:
+        # set metrics
+        if multilabel and metrics is None:
+            metrics = ["binary_accuracy"]
+        elif metrics is None:
             metrics = ["accuracy"]
+
         # set number of classes and multilabel flag
         num_classes = U.nclasses_from_data(train_data)
 
@@ -140,7 +145,7 @@ def tabular_classifier(
     name,
     train_data,
     multilabel=None,
-    metrics=["accuracy"],
+    metrics=None,
     hidden_layers=[1000, 500],
     hidden_dropouts=[0.0, 0.5],
     bn=False,
@@ -156,7 +161,8 @@ def tabular_classifier(
         multilabel (bool):  If True, multilabel model will be returned.
                             If false, binary/multiclass model will be returned.
                             If None, multilabel will be inferred from data.
-        metrics(list): list of metrics to use
+        metrics(list): List of metrics to use.  If None: 'accuracy' is used for binar/multiclass,
+                       'binary_accuracy' is used for multilabel classification, and 'mae' is used for regressio
         hidden_layers(list): number of units in each hidden layer of NN
         hidden_dropouts(list): Dropout values after each hidden layer of NN
         bn(bool): If True, BatchNormalization will be used before each fully-connected layer in NN

ktrain/text/models.py

@@ -58,7 +58,7 @@ def _text_model(
     preproc=None,
     multilabel=None,
     classification=True,
-    metrics=["accuracy"],
+    metrics=None,
     verbose=1,
 ):
     """
@@ -82,7 +82,8 @@ def _text_model(
                             If None, multilabel will be inferred from data.
         classification(bool): If True, will build a text classificaton model.
                               Otherwise, a text regression model will be returned.
-        metrics(list): list of metrics to use
+        metrics(list): List of metrics to use.  If None: 'accuracy' is used for binar/multiclassification,
+                       'binary_accuracy' is used for multilabel classification, and 'mae' is used for regression.
         verbose (boolean): verbosity of output
     Return:
         model (Model): A Keras Model instance
@@ -135,8 +136,6 @@ def _text_model(
         maxlen = U.shape_from_data(train_data)[1]
         U.vprint("maxlen is %s" % (maxlen), verbose=verbose)
     else:  # classification
-        if metrics is None:
-            metrics = ["accuracy"]
         # set number of classes and multilabel flag
         num_classes = U.nclasses_from_data(train_data)
 
@@ -151,6 +150,12 @@ def _text_model(
             name = FASTTEXT
         U.vprint("Is Multi-Label? %s" % (multilabel), verbose=verbose)
 
+        # set metrics
+        if multilabel and metrics is None:
+            metrics = ["binary_accuracy"]
+        elif metrics is None:
+            metrics = ["accuracy"]
+
         # set loss and activations
         loss_func = "categorical_crossentropy"
         activation = "softmax"
@@ -548,7 +553,7 @@ def _build_bigru(
 
 
 def text_classifier(
-    name, train_data, preproc=None, multilabel=None, metrics=["accuracy"], verbose=1
+    name, train_data, preproc=None, multilabel=None, metrics=None, verbose=1
 ):
     """
     ```
@@ -570,7 +575,8 @@ def text_classifier(
         multilabel (bool):  If True, multilabel model will be returned.
                             If false, binary/multiclass model will be returned.
                             If None, multilabel will be inferred from data.
-        metrics(list): metrics to use
+        metrics(list): List of metrics to use.  If None: 'accuracy' is used for binar/multiclassification,
+                       'binary_accuracy' is used for multilabel classification, and 'mae' is used for regression.
         verbose (boolean): verbosity of output
     Return:
         model (Model): A Keras Model instance

ktrain/vision/models.py

@@ -118,7 +118,7 @@ def image_classifier(
     train_data,
     val_data=None,
     freeze_layers=None,
-    metrics=["accuracy"],
+    metrics=None,
     optimizer_name=U.DEFAULT_OPT,
     multilabel=None,
     pt_fc=[],
@@ -140,9 +140,11 @@ def image_classifier(
                             If None, then all layers except new Dense layers
                             will be frozen/untrainable.
         metrics (list):  metrics to use
+        metrics(list): List of metrics to use.  If None: 'accuracy' is used for binar/multiclass,
+                       'binary_accuracy' is used for multilabel classification, and 'mae' is used for regression
         optimizer_name(str|obj): name of Keras optimizer (e.g., 'adam', 'sgd') or instance of keras Optimizer
         multilabel(bool):  If True, model will be build to support
-                           multilabel classificaiton (labels are not mutually exclusive).
+                           multilabel classification (labels are not mutually exclusive).
                            If False, binary/multiclassification model will be returned.
                            If None, multilabel status will be inferred from data.
         pt_fc (list of ints): number of hidden units in extra Dense layers
@@ -234,7 +236,7 @@ def image_model(
     train_data,
     val_data=None,
     freeze_layers=None,
-    metrics=["accuracy"],
+    metrics=None,
     optimizer_name=U.DEFAULT_OPT,
     multilabel=None,
     pt_fc=[],
@@ -255,7 +257,8 @@ def image_model(
         freeze_layers (int):  number of beginning layers to make untrainable
                             If None, then all layers except new Dense layers
                             will be frozen/untrainable.
-        metrics (list):  metrics to use
+        metrics(list): List of metrics to use.  If None: 'accuracy' is used for binar/multiclass,
+                       'binary_accuracy' is used for multilabel classification, and 'mae' is used for regression
         optimizer_name(str|obj): name of Keras optimizer (e.g., 'adam', 'sgd') or instance of Keras optimizer
         multilabel(bool):  If True, model will be build to support
                            multilabel classificaiton (labels are not mutually exclusive).
@@ -328,6 +331,14 @@ def image_model(
     if not multilabel and len(train_data[0][-1].shape) == 1:
         is_regression = True
 
+    # set metrics
+    if is_regression and metrics is None:
+        metrics = ["mae"]
+    elif multilabel and metrics is None:
+        metrics = ["binary_accuracy"]
+    elif metrics is None:
+        metrics = ["accuracy"]
+
     # set loss and acivations
     loss_func = "categorical_crossentropy"
     activation = "softmax"
@@ -337,8 +348,6 @@ def image_model(
     elif is_regression:
         loss_func = "mse"
         activation = None
-        if metrics == ["accuracy"]:
-            metrics = ["mae"]
 
     U.vprint("Is Multi-Label? %s" % (multilabel), verbose=verbose)
     U.vprint("Is Regression? %s" % (is_regression), verbose=verbose)