do not reset NN layers in online (streaming) mode

madmom/ml/nn/__init__.py

@@ -71,8 +71,9 @@ class NeuralNetwork(Processor):
 
     """
 
-    def __init__(self, layers):
+    def __init__(self, layers, online=False):
         self.layers = layers
+        self.online = online
 
     def process(self, data):
         """
@@ -89,13 +90,16 @@ def process(self, data):
             Network predictions for this data.
 
         """
+        # reset the layers? (online: do not reset, keep the state)
+        # Note: use getattr to be able to process old models
+        reset = not getattr(self, 'online', False)
         # check the dimensions of the data
         if data.ndim == 1:
             data = np.atleast_2d(data).T
         # loop over all layers
         for layer in self.layers:
             # activate the layer and feed the output into the next one
-            data = layer(data)
+            data = layer(data, reset=reset)
         # ravel the predictions if needed
         if data.ndim == 2 and data.shape[1] == 1:
             data = data.ravel()

madmom/ml/nn/layers.py

@@ -23,9 +23,9 @@ class Layer(object):
 
     """
 
-    def __call__(self, *args):
+    def __call__(self, *args, **kwargs):
         # this magic method makes a Layer callable
-        return self.activate(*args)
+        return self.activate(*args, **kwargs)
 
     def activate(self, data):
         """
@@ -65,7 +65,7 @@ def __init__(self, weights, bias, activation_fn):
         self.bias = bias
         self.activation_fn = activation_fn
 
-    def activate(self, data):
+    def activate(self, data, **kwargs):
         """
         Activate the layer.
 
@@ -180,7 +180,7 @@ def __init__(self, fwd_layer, bwd_layer):
         self.fwd_layer = fwd_layer
         self.bwd_layer = bwd_layer
 
-    def activate(self, data):
+    def activate(self, data, **kwargs):
         """
         Activate the layer.
 
@@ -200,9 +200,9 @@ def activate(self, data):
 
         """
         # activate in forward direction
-        fwd = self.fwd_layer(data)
+        fwd = self.fwd_layer(data, **kwargs)
         # also activate with reverse input
-        bwd = self.bwd_layer(data[::-1])
+        bwd = self.bwd_layer(data[::-1], **kwargs)
         # stack data
         return np.hstack((fwd, bwd[::-1]))
 
@@ -700,7 +700,7 @@ def __init__(self, weights, bias, stride=1, pad='valid',
             raise NotImplementedError('only `pad` == "valid" implemented.')
         self.pad = pad
 
-    def activate(self, data):
+    def activate(self, data, **kwargs):
         """
         Activate the layer.
 
@@ -757,7 +757,7 @@ class StrideLayer(Layer):
     def __init__(self, block_size):
         self.block_size = block_size
 
-    def activate(self, data):
+    def activate(self, data, **kwargs):
         """
         Activate the layer.
 
@@ -798,7 +798,7 @@ def __init__(self, size, stride=None):
             stride = size
         self.stride = stride
 
-    def activate(self, data):
+    def activate(self, data, **kwargs):
         """
         Activate the layer.
 
@@ -862,7 +862,7 @@ def __init__(self, beta, gamma, mean, inv_std, activation_fn):
         self.inv_std = inv_std
         self.activation_fn = activation_fn
 
-    def activate(self, data):
+    def activate(self, data, **kwargs):
         """
         Activate the layer.
 

madmom/processors.py

@@ -847,3 +847,4 @@ def io_arguments(parser, output_suffix='.txt', pickle=True, online=False):
         sp.set_defaults(origin='future')
         sp.set_defaults(num_frames=1)
         sp.set_defaults(stream=None)
+        sp.set_defaults(online=True)