Merge branch 'main' of github.com:tensorlayer/TensorLayerX into main

Author: Laicheng0830

docs/modules/nn.rst

@@ -48,27 +48,26 @@ Layer list
    GroupConv2d
 
    PadLayer
-   PoolLayer
    ZeroPad1d
    ZeroPad2d
    ZeroPad3d
    MaxPool1d
-   MeanPool1d
+   AvgPool1d
    MaxPool2d
-   MeanPool2d
+   AvgPool2d
    MaxPool3d
-   MeanPool3d
+   AvgPool3d
    GlobalMaxPool1d
-   GlobalMeanPool1d
+   GlobalAvgPool1d
    GlobalMaxPool2d
-   GlobalMeanPool2d
+   GlobalAvgPool2d
    GlobalMaxPool3d
-   GlobalMeanPool3d
-   AdaptiveMeanPool1d
+   GlobalAvgPool3d
+   AdaptiveAvgPool1d
    AdaptiveMaxPool1d
-   AdaptiveMeanPool2d
+   AdaptiveAvgPool2d
    AdaptiveMaxPool2d
-   AdaptiveMeanPool3d
+   AdaptiveAvgPool3d
    AdaptiveMaxPool3d
    CornerPool2d
 
@@ -410,83 +409,77 @@ Padding layer for any modes.
 Pooling Layers
 ------------------------
 
-Pool Layer (Expert API)
-^^^^^^^^^^^^^^^^^^^^^^^^^
-Pooling layer for any dimensions and any pooling functions.
-
-.. autoclass:: PoolLayer
-
 1D Max pooling
 ^^^^^^^^^^^^^^^^^^^
 .. autoclass:: MaxPool1d
 
-1D Mean pooling
+1D Avg pooling
 ^^^^^^^^^^^^^^^^^^^
-.. autoclass:: MeanPool1d
+.. autoclass:: AvgPool1d
 
 2D Max pooling
 ^^^^^^^^^^^^^^^^^^^
 .. autoclass:: MaxPool2d
 
-2D Mean pooling
+2D Avg pooling
 ^^^^^^^^^^^^^^^^^^^
-.. autoclass:: MeanPool2d
+.. autoclass:: AvgPool2d
 
 3D Max pooling
 ^^^^^^^^^^^^^^^^^^^
 .. autoclass:: MaxPool3d
 
-3D Mean pooling
+3D Avg pooling
 ^^^^^^^^^^^^^^^^^^^
-.. autoclass:: MeanPool3d
+.. autoclass:: AvgPool3d
 
 1D Global Max pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
 .. autoclass:: GlobalMaxPool1d
 
-1D Global Mean pooling
+1D Global Avg pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
-.. autoclass:: GlobalMeanPool1d
+.. autoclass:: GlobalAvgPool1d
 
 2D Global Max pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
 .. autoclass:: GlobalMaxPool2d
 
-2D Global Mean pooling
+2D Global Avg pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
-.. autoclass:: GlobalMeanPool2d
+.. autoclass:: GlobalAvgPool2d
 
 3D Global Max pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
 .. autoclass:: GlobalMaxPool3d
 
-3D Global Mean pooling
+3D Global Avg pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
-.. autoclass:: GlobalMeanPool3d
+.. autoclass:: GlobalAvgPool3d
 
 1D Adaptive Max pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
 .. autoclass:: AdaptiveMaxPool1d
 
-1D Adaptive Mean pooling
+1D Adaptive Avg pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
-.. autoclass:: AdaptiveMeanPool1d
+.. autoclass:: AdaptiveAvgPool1d
 
 2D Adaptive Max pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
 .. autoclass:: AdaptiveMaxPool2d
 
-2D Adaptive Mean pooling
+2D Adaptive Avg pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
-.. autoclass:: AdaptiveMeanPool2d
+.. autoclass:: AdaptiveAvgPool2d
 
 3D Adaptive Max pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
 .. autoclass:: AdaptiveMaxPool3d
 
-3D Adaptive Mean pooling
+3D Adaptive Avg pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
-.. autoclass:: AdaptiveMeanPool3d
+.. autoclass:: AdaptiveAvgPool3d
 
 2D Corner pooling
 ^^^^^^^^^^^^^^^^^^^^^^^^^^

tensorlayerx/backend/ops/paddle_backend.py

@@ -1211,10 +1211,10 @@ def floor(x):
 
 
 def gather(params, indices, axis=None):
-    if axis < 0:
-        axis = len(params.shape) + axis
     if axis is None:
         axis = 0
+    if axis < 0:
+        axis = len(params.shape) + axis
     return pd.gather(params, indices, axis)
 
 

tensorlayerx/backend/ops/paddle_nn.py

@@ -681,6 +681,7 @@ class MaxPool(object):
     def __init__(self, ksize, strides, padding, data_format=None):
         self.data_format, self.padding = preprocess_2d_format(data_format, padding)
         self.ksize = ksize
+        print(self.data_format)
         if self.data_format is 'NHWC':
             self._stride = (strides[1], strides[2])
         elif self.data_format is 'NCHW':

tensorlayerx/backend/ops/tensorflow_nn.py

@@ -868,8 +868,7 @@ class AvgPool(object):
     def __init__(self, ksize, strides, padding, data_format=None):
         self.ksize = ksize
         self.strides = strides
-        self.data_format = data_format
-        self.padding = padding_format(padding)
+        self.data_format, self.padding = preprocess_2d_format(data_format=data_format, padding=padding)
         self.padding_value = None
         if not isinstance(self.padding, str):
             self.padding_value = preprocess_padding(self.padding, '2d', self.data_format)
@@ -1334,7 +1333,6 @@ def __call__(self, input, filters):
                 output_w = input_w * strides_w + max(kernel_w - strides_w, 0) - 2 * self._padding
                 self.padding = [[0, 0], [self._padding, self._padding],[self._padding, self._padding], [0, 0]]
             else:
-                print(input_h, strides_h, kernel_h, strides_h, self._padding[0], self._padding)
                 output_h = input_h * strides_h + max(kernel_h - strides_h, 0) - 2 * self._padding[0]
                 output_w = input_w * strides_w + max(kernel_w - strides_w, 0) - 2* self._padding[1]
                 self.padding = [[0, 0], [self._padding[0], self._padding[0]],[self._padding[1], self._padding[1]], [0, 0]]

tensorlayerx/backend/ops/torch_backend.py

@@ -547,11 +547,13 @@ def __init__(self, axis=None, keepdims=False):
 
     def __call__(self, inputs):
         if self.axis is not None:
-            out = inputs
-            for dim in self.axis[-1]:
-                print(dim)
-                out = torch.max(out, dim=dim, keepdim=self.keepdims).values
-            return out
+            if isinstance(self.axis, (list, tuple)):
+                out = inputs
+                for dim in self.axis[::-1]:
+                    out = torch.max(out, dim=dim, keepdim=self.keepdims).values
+                return out
+            else:
+                return torch.max(inputs, dim=self.axis, keepdim=self.keepdims).values
         else:
             return torch.max(inputs)
 

tensorlayerx/backend/ops/torch_nn.py

@@ -792,6 +792,8 @@ def __init__(self, ksize, strides, padding, data_format=None):
         elif data_format in ['channels_first', 'NCL', 'NCW', 'NCHW', 'NCDHW']:
             self.data_format = 'channels_first'
         self.padding = padding
+        if self.padding in ['VALID', 'valid']:
+            self.padding = 0
 
     def __call__(self, inputs):
         if self.data_format == 'channels_last':
@@ -818,7 +820,7 @@ def __call__(self, inputs):
             return out
 
     def maxpool1d_same_padding(self, input):
-        rows_odd, padding_rows = same_padding(input, self.ksize, self.strides, 1)
+        rows_odd, padding_rows = same_padding(input, self.ksize, self.strides[0], 1)
         if rows_odd:
             input = F.pad(input, [0, int(rows_odd)], 'constant', float('-inf'))
         return F.max_pool1d(input, self.ksize, self.strides, padding=(padding_rows // 2))
@@ -894,6 +896,8 @@ def __init__(self, ksize, strides, padding, data_format=None):
         elif data_format in ['channels_first', 'NCL', 'NCW', 'NCHW', 'NCDHW']:
             self.data_format = 'channels_first'
         self.padding = padding
+        if self.padding in ['VALID', 'valid']:
+            self.padding = 0
 
     def __call__(self, inputs):
         if self.data_format == 'channels_last':
@@ -920,7 +924,7 @@ def __call__(self, inputs):
             return out
 
     def avgpool1d_same_padding(self, input):
-        rows_odd, padding_rows = same_padding(input, self.ksize, self.strides, 1)
+        rows_odd, padding_rows = same_padding(input, self.ksize, self.strides[0], 1)
         if rows_odd:
             input = F.pad(input, [0, int(rows_odd)], 'replicate')
         return F.avg_pool1d(input, self.ksize, self.strides, padding=(padding_rows // 2))

tensorlayerx/nn/core/core_tensorflow.py

@@ -1118,8 +1118,6 @@ class ParameterDict(Module):
     ---------
     __init__()
         Initializing the ParameterDict.
-    setdefault()
-
     clear()
         Remove all items from the ParameterDict.
     setdefault(key, default=None)