Merge branch 'develop'

CHANGELOG.md

@@ -7,6 +7,19 @@ Most recent releases are shown at the top. Each release shows:
 - **Fixed**: Bug fixes that don't change documented behaviour
 
 
+## 0.16.3 (2020-06-10)
+
+### New:
+- added `metrics` parameter to `text.text_classifier` and `text.text_regression_model` functions
+- added `metrics` parameter to `Transformer.get_classifier` and `Transformer.get_regrssion_model` methods
+
+### Changed
+- `metric` parameter in `vision.image_classifier` and `vision.image_regression_model` functions changed to `metrics`
+
+### Fixed:
+- N/A
+
+
 ## 0.16.2 (2020-06-07)
 
 ### New:

ktrain/text/models.py

@@ -52,7 +52,7 @@ def calc_r(y_i, x, y):
     return np.log(calc_pr(y_i, x, y, True) / calc_pr(y_i, x, y, False))
 
 
-def _text_model(name, train_data, preproc=None, multilabel=None, classification=True, verbose=1):
+def _text_model(name, train_data, preproc=None, multilabel=None, classification=True, metrics=['accuracy'], verbose=1):
     """
     Build and return a text classification or text regression model.
 
@@ -73,6 +73,7 @@ def _text_model(name, train_data, preproc=None, multilabel=None, classification=
                             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
         verbose (boolean): verbosity of output
     Return:
         model (Model): A Keras Model instance
@@ -103,7 +104,7 @@ def _text_model(name, train_data, preproc=None, multilabel=None, classification=
         raise ValueError('you preprocessed for %s but want to build a %s model' % (preproc.name, name))
 
     if not classification: # regression
-        metrics=['mae']
+        if metrics is None or metrics==['accuracy']: metrics=['mae']
         num_classes = 1
         multilabel = False
         loss_func = 'mse'
@@ -113,7 +114,7 @@ def _text_model(name, train_data, preproc=None, multilabel=None, classification=
         maxlen = U.shape_from_data(train_data)[1]
         U.vprint('maxlen is %s' % (maxlen), verbose=verbose)
     else:                 # classification
-        metrics = ['accuracy']
+        if metrics is None: metrics = ['accuracy']
         # set number of classes and multilabel flag
         num_classes = U.nclasses_from_data(train_data)
 
@@ -436,7 +437,7 @@ def _build_bigru(num_classes,
 
 
 
-def text_classifier(name, train_data, preproc=None, multilabel=None, verbose=1):
+def text_classifier(name, train_data, preproc=None, multilabel=None, metrics=['accuracy'], verbose=1):
     """
     Build and return a text classification model.
 
@@ -456,6 +457,7 @@ def text_classifier(name, train_data, preproc=None, multilabel=None, verbose=1):
         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
         verbose (boolean): verbosity of output
     Return:
         model (Model): A Keras Model instance
@@ -465,10 +467,10 @@ def text_classifier(name, train_data, preproc=None, multilabel=None, verbose=1):
     if preproc is not None and not preproc.get_classes():
         raise ValueError('preproc.get_classes() is empty, but required for text classification')
     return _text_model(name, train_data, preproc=preproc,
-                       multilabel=multilabel, classification=True, verbose=verbose)
+                       multilabel=multilabel, classification=True, metrics=metrics, verbose=verbose)
 
 
-def text_regression_model(name, train_data, preproc=None, verbose=1):
+def text_regression_model(name, train_data, preproc=None, metrics=['mae'],  verbose=1):
     """
     Build and return a text regression model.
 
@@ -484,6 +486,7 @@ def text_regression_model(name, train_data, preproc=None, verbose=1):
         train_data (tuple): a tuple of numpy.ndarrays: (x_train, y_train)
         preproc: a ktrain.text.TextPreprocessor instance.
                  As of v0.8.0, this is required.
+        metrics(list): metrics to use
         verbose (boolean): verbosity of output
     Return:
         model (Model): A Keras Model instance
@@ -493,4 +496,4 @@ def text_regression_model(name, train_data, preproc=None, verbose=1):
     if preproc is not None and preproc.get_classes():
         raise ValueError('preproc.get_classes() is supposed to be empty for text regression tasks')
     return _text_model(name, train_data, preproc=preproc,
-                      multilabel=False, classification=False,  verbose=verbose)
+                      multilabel=False, classification=False, metrics=metrics, verbose=verbose)

ktrain/text/preprocessor.py

@@ -919,15 +919,16 @@ def _load_pretrained(self, mname, num_labels):
 
 
 
-    def get_classifier(self, fpath=None, multilabel=None):
+    def get_classifier(self, fpath=None, multilabel=None, metrics=['accuracy']):
         """
-        creates a model for classification
+        creates a model for text classification
         Args:
           fpath(str): optional path to saved pretrained model. Typically left as None.
           multilabel(bool): If None, multilabel status is discovered from data [recommended].
                             If True, model will be forcibly configured for multilabel task.
                             If False, model will be forcibly configured for non-multilabel task.
                             It is recommended to leave this as None.
+          metrics(list): metrics to use
         """
         self.check_trained()
         if not self.get_classes():
@@ -953,11 +954,17 @@ def get_classifier(self, fpath=None, multilabel=None):
             loss_fn = keras.losses.CategoricalCrossentropy(from_logits=True)
         model.compile(loss=loss_fn,
                       optimizer=U.DEFAULT_OPT,
-                      metrics=['accuracy'])
+                      metrics=metrics)
         return model
 
 
-    def get_regression_model(self, fpath=None):
+    def get_regression_model(self, fpath=None, metrics=['mae']):
+        """
+        creates a model for text regression
+        Args:
+          fpath(str): optional path to saved pretrained model. Typically left as None.
+          metrics(list): metrics to use
+        """
         self.check_trained()
         if self.get_classes():
             warnings.warn('class labels were provided - treating as classification problem')
@@ -968,7 +975,7 @@ def get_regression_model(self, fpath=None):
         loss_fn = 'mse'
         model.compile(loss=loss_fn,
                       optimizer=U.DEFAULT_OPT,
-                      metrics=['mae'])
+                      metrics=metrics)
         return model
 
 

ktrain/version.py

@@ -1,2 +1,2 @@
 __all__ = ['__version__']
-__version__ = '0.16.2'
+__version__ = '0.16.3'

ktrain/vision/models.py

@@ -74,7 +74,7 @@ def image_classifier(name,
                      train_data,
                      val_data=None,
                      freeze_layers=None, 
-                     metric='accuracy',
+                     metrics=['accuracy'],
                      optimizer_name = U.DEFAULT_OPT,
                      multilabel=None,
                      multigpu_number=None, 
@@ -95,7 +95,7 @@ def image_classifier(name,
         freeze_layers (int):  number of beginning layers to make untrainable
                             If None, then all layers except new Dense layers
                             will be frozen/untrainable.
-        metric (string):  metric to use
+        metrics (list):  metrics to use
         optimizer_name(str): name of Keras optimizer (e.g., 'adam', 'sgd')
         multilabel(bool):  If True, model will be build to support
                            multilabel classificaiton (labels are not mutually exclusive).
@@ -118,7 +118,7 @@ def image_classifier(name,
         
     """
     return image_model(name, train_data, val_data=val_data, freeze_layers=freeze_layers,
-                       metric=metric, optimizer_name=optimizer_name, multilabel=multilabel,
+                       metrics=metrics, optimizer_name=optimizer_name, multilabel=multilabel,
                        multigpu_number=multigpu_number,
                        pt_fc=pt_fc, pt_ps=pt_ps, verbose=verbose)
 
@@ -129,7 +129,7 @@ def image_regression_model(name,
                           train_data,
                           val_data=None,
                           freeze_layers=None, 
-                          metric='mae',
+                          metrics=['mae'],
                           optimizer_name = U.DEFAULT_OPT,
                           multigpu_number=None, 
                           pt_fc = [],
@@ -149,7 +149,7 @@ def image_regression_model(name,
         freeze_layers (int):  number of beginning layers to make untrainable
                             If None, then all layers except new Dense layers
                             will be frozen/untrainable.
-        metric (string):  metric to use
+        metrics (list):  metrics to use
         optimizer_name(str): name of Keras optimizer (e.g., 'adam', 'sgd')
         multilabel(bool):  If True, model will be build to support
                            multilabel classificaiton (labels are not mutually exclusive).
@@ -174,7 +174,7 @@ def image_regression_model(name,
 
 
     return image_model(name, train_data, val_data=val_data, freeze_layers=freeze_layers,
-                       metric=metric, optimizer_name=optimizer_name, multilabel=False,
+                       metrics=metrics, optimizer_name=optimizer_name, multilabel=False,
                        multigpu_number=multigpu_number,
                        pt_fc=pt_fc, pt_ps=pt_ps, verbose=verbose)
 
@@ -184,7 +184,7 @@ def image_model( name,
                  train_data,
                  val_data=None,
                  freeze_layers=None, 
-                 metric='accuracy',
+                 metrics=['accuracy'],
                  optimizer_name = U.DEFAULT_OPT,
                  multilabel=None,
                  multigpu_number=None, 
@@ -205,7 +205,7 @@ def image_model( name,
         freeze_layers (int):  number of beginning layers to make untrainable
                             If None, then all layers except new Dense layers
                             will be frozen/untrainable.
-        metric (string):  metric to use
+        metrics (list):  metrics to use
         optimizer_name(str): name of Keras optimizer (e.g., 'adam', 'sgd')
         multilabel(bool):  If True, model will be build to support
                            multilabel classificaiton (labels are not mutually exclusive).
@@ -278,7 +278,7 @@ def image_model( name,
     elif is_regression:
         loss_func = 'mse'
         activation = None
-        if metric == 'accuracy': metric = 'mae'
+        if metrics == ['accuracy']: metrics = ['mae']
 
     U.vprint("Is Multi-Label? %s" % (multilabel), verbose=verbose)
     U.vprint("Is Regression? %s" % (is_regression), verbose=verbose)
@@ -306,7 +306,7 @@ def image_model( name,
                                       pt_ps = pt_ps)
         parallel_model = multi_gpu_model(model, gpus=multigpu_number)
         parallel_model.compile(optimizer=optimizer_name, 
-                               loss='categorical_crossentropy', metrics=[metric])
+                               loss='categorical_crossentropy', metrics=metrics)
         return parallel_model
     else:
         model = build_visionmodel(name,
@@ -316,7 +316,7 @@ def image_model( name,
                                   activation=activation,
                                   pt_fc = pt_fc,
                                   pt_ps = pt_ps)
-        model.compile(optimizer=optimizer_name, loss=loss_func, metrics=[metric])
+        model.compile(optimizer=optimizer_name, loss=loss_func, metrics=metrics)
         return model