tensorflow cosine similarity

[beijinggao]

https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/tutorials/word2vec/word2vec_basic.py

[beijinggao]

optimizer = tf.train.GradientDescentOptimizer(1.0).minimize(loss)

 Compute the cosine similarity between minibatch examples and all embeddings.

norm = tf.sqrt(tf.reduce_sum(tf.square(embeddings), 1, keep_dims=True)) normalized_embeddings = embeddings / norm valid_embeddings = tf.nn.embedding_lookup( normalized_embeddings, valid_dataset) similarity = tf.matmul(valid_embeddings, normalized_embeddings, transpose_b=True)