I am trying to calculate the cosine similarity of wordsim set using the Elmo model. This may not make sense since it is designed for sentence word embedding, but I want to see how the model performs in the situations like these. The Elmo I am using is from: https://tfhub.dev/google/elmo/3 If I run the following code (it is modified from the documentation page to comply with TF 2.0), it will generate the tensor representation of the word. import tensorflow_hub as hub import tensorflow as tf elmo = hub.load("https://tfhub.dev/google/elmo/3") tensor_of_strings = tf.constant(["Gray", "Quick", "Lazy"]) elmo.signatures['default'](tensor_of_strings) If I try to calculate cosine similarity directly I will get the error, NotImplementedError: Cannot convert a symbolic Tensor (strided_slice_59:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported . I am not sure how to convert the Tensor to the numpy array directly, or is there a better evaluator for tensors instead of cosine similarity? Edit: This is what I did for calculate cosine similarity def cos_sim(a, b): return np.inner(a, b) / (np.linalg.norm(a) * (np.linalg.norm(b))) print("ELMo:", cos_sim(elmo.signatures['default'](tensor_of_strings)['word_emb'][0], elmo.signatures['default'](tensor_of_strings)['word_emb'][1])) Continue reading...
