Add files via upload

nn/Pooling.py

@@ -0,0 +1,86 @@
+import copy
+
+import numpy as np
+
+from Test_Parameter import Test_Parameter
+from Module import Module
+
+
+class Pooling(Module):
+    def __init__(self, row, col, poolsize, poolstride, mode="max"):
+        self.parameter = Test_Parameter((row, col))
+        self.inputs = []
+        self.data = None
+        self.poolsize = poolsize
+        self.poolstride = poolstride
+        self.mode = mode
+        self.maxpos = []
+
+    def forward(self, x):
+        self.inputs.append(x)
+        if self.mode == "max":
+            self.data = self.pooling()
+        return self
+
+    def backward(self, grad):
+        temp = copy.copy(self.inputs)
+        in_row, in_col = np.shape(self.inputs)
+        for i in range(0, in_row):
+            for j in range(0, in_col):
+                if self.maxpos.count((i, j)) == 0:
+                    temp[i][j] = 0
+        self.parameter.gradient = temp
+        if isinstance(self.inputs[0], Module):
+            self.inputs[0].backward(self.parameter.gradient)
+
+    def __call__(self, x):
+        return self.forward(x)
+
+    def pooling(self):
+        """INPUTS:
+                  inputMap - input array of the pooling layer
+                  poolSize - X-size(equivalent to Y-size) of receptive field
+                  poolStride - the stride size between successive pooling squares
+
+           OUTPUTS:
+                   outputMap - output array of the pooling layer
+
+           Padding mode - 'edge'
+        """
+        # inputMap sizes
+        in_row, in_col = np.shape(self.inputs)
+
+        # outputMap sizes
+        out_row, out_col = int(np.floor(in_row / self.poolstride)), int(np.floor(in_col / self.poolstride))
+        row_remainder, col_remainder = np.mod(in_row, self.poolstride), np.mod(in_col, self.poolstride)
+        if row_remainder != 0:
+            out_row += 1
+        if col_remainder != 0:
+            out_col += 1
+        outputMap = np.zeros((out_row, out_col))
+
+        # padding
+        temp_map = np.lib.pad(self.inputs, ((0, self.poolsize - row_remainder), (0, self.poolsize - col_remainder)), 'edge')
+
+        # max pooling
+        if self.mode == "max":
+            for r_idx in range(0, out_row):
+                for c_idx in range(0, out_col):
+                    startX = c_idx * self.poolstride
+                    startY = r_idx * self.poolstride
+                    poolField = temp_map[startY:startY + self.poolsize, startX:startX + self.poolsize]
+                    poolOut = np.max(poolField)
+                    temp = np.where(self.inputs == poolOut)
+                    self.maxpos.append((temp[0][0], temp[1][0]))
+                    outputMap[r_idx, c_idx] = poolOut
+        elif self.mode == "mean":
+            for r_idx in range(0, out_row):
+                for c_idx in range(0, out_col):
+                    startX = c_idx * self.poolstride
+                    startY = r_idx * self.poolstride
+                    poolField = temp_map[startY:startY + self.poolsize, startX:startX + self.poolsize]
+                    poolOut = np.sum(poolField)
+                    outputMap[r_idx, c_idx] = poolOut / self.poolsize / self.poolsize
+
+        # retrun outputMap
+        return outputMap