Method embed

examples/SimilarityLearning/mnist-embeddings.py:200–224

Embed all given tensors into an nfeatures-dim space.

(self, x, nfeatures=2)

class EmbeddingModel(ModelDesc):
    def embed(self, x, nfeatures=2):
        """Embed all given tensors into an nfeatures-dim space.  """
        list_split = 0
        if isinstance(x, list):
            list_split = len(x)
            x = tf.concat(x, 0)

        # pre-process MNIST dataflow data
        x = tf.expand_dims(x, 3)
        x = x * 2 - 1

        # the embedding network
        net = slim.layers.conv2d(x, 20, 5, scope='conv1')
        net = slim.layers.max_pool2d(net, 2, scope='pool1')
        net = slim.layers.conv2d(net, 50, 5, scope='conv2')
        net = slim.layers.max_pool2d(net, 2, scope='pool2')
        net = slim.layers.flatten(net, scope='flatten3')
        net = slim.layers.fully_connected(net, 500, scope='fully_connected4')
        embeddings = slim.layers.fully_connected(net, nfeatures, activation_fn=None, scope='fully_connected5')

        # if "x" was a list of tensors, then split the embeddings
        if list_split > 0:
            embeddings = tf.split(embeddings, list_split, 0)

        return embeddings

    def optimizer(self):
        lr = tf.get_variable('learning_rate', initializer=1e-4, trainable=False)