Add embedding layer.

Author: lucasb-eyer

DeepFried2/layers/Embedding.py

@@ -0,0 +1,21 @@
+import DeepFried2 as df
+
+
+class Embedding(df.Module):
+    def __init__(self, ntok, ndim, init=df.init.ortho_svd()):
+        """A layer that learns `ntok` embedding vectors of dimension `ndim`.
+
+        Note that this doesn't take care of any `unk`/`oov` token.
+        If you need that, increase `ntok` by one and handle it from the outside.
+
+        The input to this layer is an array of indices of any arbitrary shape.
+        The output will be an array of the same shape plus an added dimension
+        for the embeddings at the end (as last dimension).
+        """
+        df.Module.__init__(self)
+
+        self.ndim = ndim
+        self.W = self._addparam((ntok, ndim), init, name='Wemb_{}x{}'.format(ntok, ndim))
+
+    def symb_forward(self, symb_input):
+        return self.W.param[symb_input.flatten()].reshape(tuple(symb_input.shape) + (self.ndim,))

DeepFried2/layers/__init__.py

@@ -1,9 +1,10 @@
 from .AddConstant import AddConstant
 from .BatchNormalization import BatchNormalization
+from .Bias import Bias
 from .Dropout import Dropout
+from .Embedding import Embedding
 from .Identity import Identity
 from .Linear import Linear
-from .Bias import Bias
 from .Log import Log
 from .ReLU import ReLU
 from .ELU import ELU

DeepFried2/tests/layers/test_Embedding.py

@@ -0,0 +1,22 @@
+#!/usr/bin/env python3
+
+import DeepFried2 as df
+
+import unittest
+import numpy as np
+
+class TestEmbedding(unittest.TestCase):
+
+    def testForward(self):
+        X = np.array([
+            [0, 1, 2],
+            [2, 1, 0],
+        ])
+
+        Z = np.array([
+            [[1,0,0,0], [0,1,0,0], [0,0,1,0]],
+            [[0,0,1,0], [0,1,0,0], [1,0,0,0]],
+        ])
+
+        Y = df.Embedding(ntok=3, ndim=4, init=df.init.eye()).forward(X)
+        np.testing.assert_array_equal(Y, Z)