Fixed MXNet interface (#42)

* added aux_states to mxnet

* fix coverage

* fixed overintendation

Author: wielandbrendel

foolbox/models/mxnet.py

@@ -14,9 +14,9 @@ class MXNetModel(DifferentiableModel):
         The input to the model.
     logits : `mxnet.symbol.Symbol`
         The predictions of the model, before the softmax.
-    weights : `dictionary mapping str to mxnet.nd.array`
-        The weights of the model.
-    device : `mxnet.context.Context`
+    args : `dictionary mapping str to mxnet.nd.array`
+        The parameters of the model.
+    ctx : `mxnet.context.Context`
         The device, e.g. mxnet.cpu() or mxnet.gpu().
     num_classes : int
         The number of classes.
@@ -25,6 +25,8 @@ class MXNetModel(DifferentiableModel):
         (0, 1) or (0, 255).
     channel_axis : int
         The index of the axis that represents color channels.
+    aux_states : `dictionary mapping str to mxnet.nd.array`
+        The states of auxiliary parameters of the model.
     preprocessing: 2-element tuple with floats or numpy arrays
         Elementwises preprocessing of input; we first subtract the first
         element of preprocessing from the input and then divide the input by
@@ -36,11 +38,12 @@ def __init__(
             self,
             data,
             logits,
-            weights,
-            device,
+            args,
+            ctx,
             num_classes,
             bounds,
             channel_axis=1,
+            aux_states=None,
             preprocessing=(0, 1)):
 
         super(MXNetModel, self).__init__(
@@ -52,7 +55,7 @@ def __init__(
 
         self._num_classes = num_classes
 
-        self._device = device
+        self._device = ctx
 
         self._data_sym = data
         self._batch_logits_sym = logits
@@ -63,9 +66,18 @@ def __init__(
         loss = mx.symbol.softmax_cross_entropy(logits, label)
         self._loss_sym = loss
 
-        weight_names = list(weights.keys())
-        weight_arrays = [weights[name] for name in weight_names]
-        self._args_map = dict(zip(weight_names, weight_arrays))
+        self._args_map = args.copy()
+        self._aux_map = aux_states.copy() if aux_states is not None else None
+
+        # move all parameters to correct device
+        for k in self._args_map.keys():
+            self._args_map[k] = \
+                self._args_map[k].as_in_context(ctx)      # pragma: no cover
+
+        if aux_states is not None:
+            for k in self._aux_map.keys():                # pragma: no cover
+                self._aux_map[k] = \
+                    self._aux_map[k].as_in_context(ctx)   # pragma: no cover
 
     def num_classes(self):
         return self._num_classes
@@ -76,7 +88,8 @@ def batch_predictions(self, images):
         data_array = mx.nd.array(images, ctx=self._device)
         self._args_map[self._data_sym.name] = data_array
         model = self._batch_logits_sym.bind(
-            ctx=self._device, args=self._args_map, grad_req='null')
+            ctx=self._device, args=self._args_map, grad_req='null',
+            aux_states=self._aux_map)
         model.forward(is_train=False)
         logits_array = model.outputs[0]
         logits = logits_array.asnumpy()
@@ -99,7 +112,8 @@ def predictions_and_gradient(self, image, label):
             ctx=self._device,
             args=self._args_map,
             args_grad=grad_map,
-            grad_req='write')
+            grad_req='write',
+            aux_states=self._aux_map)
         model.forward(is_train=True)
         logits_array = model.outputs[0]
         model.backward([
@@ -119,7 +133,8 @@ def _loss_fn(self, image, label):
         self._args_map[self._data_sym.name] = data_array
         self._args_map[self._label_sym.name] = label_array
         model = self._loss_sym.bind(
-            ctx=self._device, args=self._args_map, grad_req='null')
+            ctx=self._device, args=self._args_map, grad_req='null',
+            aux_states=self._aux_map)
         model.forward(is_train=False)
         loss_array = model.outputs[0]
         loss = loss_array.asnumpy()[0]

foolbox/tests/test_models_mxnet.py

@@ -21,7 +21,7 @@ def mean_brightness_net(images):
         images,
         logits,
         {},
-        device=mx.cpu(),
+        ctx=mx.cpu(),
         num_classes=num_classes,
         bounds=bounds,
         channel_axis=1)
@@ -68,7 +68,7 @@ def mean_brightness_net(images):
         images,
         logits,
         {},
-        device=mx.cpu(),
+        ctx=mx.cpu(),
         num_classes=num_classes,
         bounds=bounds,
         preprocessing=preprocessing,