Trainer refactoring

delira/__init__.py

@@ -65,6 +65,11 @@ def get_backends():
     """
     Return List of currently available backends
 
+    Returns
+    -------
+    list
+        list of strings containing the currently installed backends
+
     """
 
     if not __BACKENDS:

delira/data_loading/dataset.py

@@ -154,7 +154,7 @@ def train_test_split(self, *args, **kwargs):
         """
         split dataset into train and test data
 
-        .. deprecated:: 0.3
+        .. deprecated-removed:: 0.3 0.4
             method will be removed in next major release
 
         Parameters
@@ -692,8 +692,10 @@ def _make_dataset(self, path):
 
 
 if "TORCH" in get_backends():
+
     from torchvision.datasets import CIFAR10, CIFAR100, EMNIST, MNIST, \
         FashionMNIST
+    import torch
 
     class TorchvisionClassificationDataset(AbstractDataset):
         """
@@ -802,8 +804,15 @@ def __getitem__(self, index):
             """
 
             data = self.data[index]
+            label = data[1]
+
+            if isinstance(label, torch.Tensor):
+                label = label.numpy()
+            elif isinstance(label, int):
+                label = np.array(label)
             data_dict = {"data": np.array(data[0]),
-                         "label": data[1].reshape(1).astype(np.float32)}
+
+                         "label": label.reshape(1).astype(np.float32)}
 
             if self.one_hot:
                 # TODO: Remove and refer to batchgenerators transform:

delira/data_loading/load_utils.py

@@ -67,11 +67,14 @@ def norm_zero_mean_unit_std(data):
 
 
 @make_deprecated("LoadSample")
-def _is_valid_image_file(fname, img_extensions, gt_extensions):
+def is_valid_image_file(fname, img_extensions, gt_extensions):
     """
     Helper Function to check wheter file is image file and has at least
     one label file
-     Parameters
+
+    .. deprecated-removed:: 0.3.4 0.3.5
+
+    Parameters
     ----------
     fname : str
         filename of image path
@@ -99,6 +102,8 @@ def default_load_fn_2d(img_file, *label_files, img_shape, n_channels=1):
     """
     loading single 2d sample with arbitrary number of samples
 
+    .. deprecated-removed:: 0.3.4 0.3.5
+
     Parameters
     ----------
     img_file : string
@@ -136,14 +141,16 @@ def default_load_fn_2d(img_file, *label_files, img_shape, n_channels=1):
 
 
 class LoadSample:
+    """
+    Provides a callable to load a single sample from multiple files in a folder
+    """
+
     def __init__(self,
                  sample_ext: dict,
                  sample_fn: collections.abc.Callable,
                  dtype={}, normalize=(), norm_fn=norm_range('-1,1'),
                  **kwargs):
         """
-        Provides a callable to load a single sample from multiple files in a
-        folder
 
         Parameters
         ----------

delira/data_loading/nii.py

@@ -32,6 +32,8 @@ def load_sample_nii(files, label_load_cls):
     """
     Load sample from multiple ITK files
 
+    .. deprecated-removed:: 0.3.4 0.4
+
     Parameters
     ----------
     files : dict with keys `img` and `label`
@@ -72,6 +74,8 @@ class BaseLabelGenerator(object):
     """
     Base Class to load labels from json files
 
+    .. deprecated-removed: 0.3.3 0.3.5
+
     """
 
     @make_deprecated('dict containing labels')

delira/logging/__init__.py

@@ -1,2 +1,3 @@
 from .multistream_handler import MultiStreamHandler
-from .trixi_handler import TrixiHandler
+from .trixi_handler import TrixiHandler, VisdomLoggingHandler, \
+    TensorboardXLoggingHandler

delira/logging/trixi_handler.py

@@ -80,3 +80,59 @@ def emit(self, record):
                         kwargs = {}
 
                     getattr(self._logger, _prefix + key)(*args, **kwargs)
+
+
+class TensorboardXLoggingHandler(TrixiHandler):
+    """
+    Logging Handler to log with TensorboardX (via Trixi)
+    """
+
+    def __init__(self, log_dir, level=NOTSET, **kwargs):
+        """
+
+        Parameters
+        ----------
+        log_dir : str
+            path to log to
+        level : int (default: NOTSET)
+            logging level
+        **kwargs :
+            additional keyword arguments
+
+        """
+
+        from trixi.logger.tensorboard import TensorboardXLogger
+
+        super().__init__(TensorboardXLogger, level=level,
+                         target_dir=log_dir, **kwargs)
+
+
+class VisdomLoggingHandler(TrixiHandler):
+    """
+    Logging Handler to log with Visdom (via Trixi)
+
+    """
+
+    def __init__(self, exp_name, server="http://localhost", port=8080,
+                 level=NOTSET, **kwargs):
+        """
+
+        Parameters
+        ----------
+        exp_name : str
+            experiment name
+        server : str
+            address of visdom server
+        port : int
+            port of visdom server
+        level : int (default: NOTSET)
+            logging level
+        **kwargs :
+            additional keyword arguments
+
+        """
+
+        from trixi.logger.visdom import NumpyVisdomLogger
+
+        super().__init__(NumpyVisdomLogger, level=level, exp_name=exp_name,
+                         server=server, port=port, **kwargs)

delira/models/abstract_network.py

@@ -53,7 +53,7 @@ def __call__(self, *args, **kwargs):
 
     @staticmethod
     @abc.abstractmethod
-    def closure(model, data_dict: dict, optimizers: dict, criterions={},
+    def closure(model, data_dict: dict, optimizers: dict, losses={},
                 metrics={}, fold=0, **kwargs):
         """
         Function which handles prediction from batch, logging, loss calculation
@@ -66,8 +66,8 @@ def closure(model, data_dict: dict, optimizers: dict, criterions={},
             dictionary containing the data
         optimizers : dict
             dictionary containing all optimizers to perform parameter update
-        criterions : dict
-            Functions or classes to calculate criterions
+        losses : dict
+            Functions or classes to calculate losses
         metrics : dict
             Functions or classes to calculate other metrics
         fold : int
@@ -80,8 +80,8 @@ def closure(model, data_dict: dict, optimizers: dict, criterions={},
         dict
             Metric values (with same keys as input dict metrics)
         dict
-            Loss values (with same keys as input dict criterions)
-        list
+            Loss values (with same keys as input dict losses)
+        dict
             Arbitrary number of predictions
 
         Raises
@@ -256,21 +256,23 @@ def __init__(self, sess=tf.Session, **kwargs):
             """
             AbstractNetwork.__init__(self, **kwargs)
             self._sess = sess()
-            self.inputs = None
-            self.outputs_train = None
-            self.outputs_eval = None
+            self.inputs = {}
+            self.outputs_train = {}
+            self.outputs_eval = {}
             self._losses = None
             self._optims = None
             self.training = True
 
-        def __call__(self, *args):
+        def __call__(self, *args, **kwargs):
             """
             Wrapper for calling self.run in eval setting
 
             Parameters
             ----------
             *args :
                 positional arguments (passed to `self.run`)
+            **kwargs:
+                keyword arguments (passed to `self.run`)
 
             Returns
             -------
@@ -279,7 +281,7 @@ def __call__(self, *args):
 
             """
             self.training = False
-            return self.run(*args)
+            return self.run(*args, **kwargs)
 
         def _add_losses(self, losses: dict):
             """
@@ -304,27 +306,33 @@ def _add_optims(self, optims: dict):
             """
             raise NotImplementedError()
 
-        def run(self, *args):
+        def run(self, *args, **kwargs):
             """
             Evaluates `self.outputs_train` or `self.outputs_eval` based on
             `self.training`
 
             Parameters
             ----------
             *args :
-                arguments to feed as `self.inputs`. Must have same length as
-                `self.inputs`
+                currently unused, exist for compatibility reasons
+            **kwargs :
+                kwargs used to feed as ``self.inputs``. Same keys as for
+                ``self.inputs`` must be used
 
             Returns
             -------
-            np.ndarray or list
-                based on len(self.outputs*), returns either list or np.ndarray
+            dict
+                sames keys as outputs_train or outputs_eval,
+                containing evaluated expressions as values
 
             """
-            if isinstance(self.inputs, tf.Tensor):
-                _feed_dict = dict(zip([self.inputs], args))
-            else:
-                _feed_dict = dict(zip(self.inputs, args))
+            _feed_dict = {}
+
+            for feed_key, feed_value in kwargs.items():
+                assert feed_key in self.inputs.keys(), \
+                    "{} not found in self.inputs".format(feed_key)
+                _feed_dict[self.inputs[feed_key]] = feed_value
+
             if self.training:
                 return self._sess.run(self.outputs_train, feed_dict=_feed_dict)
             else:

delira/models/classification/ResNet18.py

@@ -1,4 +1,5 @@
 import tensorflow as tf
+from delira.utils.decorators import make_deprecated
 
 conv2d = tf.keras.layers.Conv2D
 maxpool2d = tf.keras.layers.MaxPool2D
@@ -81,7 +82,9 @@ def call(self, inputs, training=None):
 
 
 class ResNet18(tf.keras.Model):
+    @make_deprecated("own repository to be announced")
     def __init__(self, num_classes=None, bias=False):
+
         super(ResNet18, self).__init__()
 
         _image_format, _axis = get_image_format_and_axis()