def test_check_output(self): ids = self.inputs['Ids'] flatten_idx = ids.flatten() padding_idx = np.random.choice(flatten_idx, 1)[0] self.outputs['Out'][np.squeeze(ids == padding_idx)] = np.zeros(31) self.attrs = {'padding_idx': cpt.long_type(padding_idx)} self.check_output()
embedding_dim, padding_idx=int(padding_idx), vocab_embeddings=table, compare=False) # print('outputs {}'.format(outputs)) # ids = np.random.randint(low=0, high=vocab_size, size=(2, 4, 5)).astype("int32") embedding("embedding_tensorIds", ids, vocab_size, embedding_dim, vocab_embeddings=table, compare=False) # ids = np.random.randint(low=0, high=vocab_size, size=(2, 4, 5)).astype("int32") flatten_idx = ids.flatten() padding_idx = np.random.choice(flatten_idx, 1)[0] # print('padding_idx {}'.format(padding_idx)) outputs = embedding("embedding_tensorIds_paddings", ids, vocab_size, embedding_dim, padding_idx=cpt.long_type(padding_idx), vocab_embeddings=table, compare=False) # print('outputs {}'.format(outputs))
embedding("embedding_none_weight", ids, vocab_size, embedding_dim, compare=False) # ids = np.random.randint(0, vocab_size, 4).astype("int32") ids = np.squeeze(ids) padding_idx = np.random.choice(ids, 1)[0] # print('padding_idx {}, ids {}'.format(padding_idx, ids)) outputs = embedding("embedding_paddings", ids, vocab_size, embedding_dim, padding_idx=int(padding_idx), vocab_embeddings=table, compare=False) # print('outputs {}'.format(outputs)) # corner case ids = np.random.randint(0, vocab_size, 4).astype("int32") pick = np.random.choice(4, 1)[0] # pick randomly to be max vacab_size -1 ids[pick] = vocab_size-1 padding_idx = -1 # print('padding_idx {}, ids {}'.format(padding_idx, ids)) outputs = embedding("embedding_paddings_neg1", ids, vocab_size, embedding_dim, padding_idx=int(padding_idx), vocab_embeddings=table, compare=False) # print('outputs {}'.format(outputs)) # ids = np.random.randint(low=0, high=vocab_size, size=(2, 4, 5)).astype("int32") embedding("embedding_tensorIds", ids, vocab_size, embedding_dim, vocab_embeddings=table, compare=False) # ids = np.random.randint(low=0, high=vocab_size, size=(2, 4, 5)).astype("int32") flatten_idx = ids.flatten() padding_idx = np.random.choice(flatten_idx, 1)[0] # print('padding_idx {}'.format(padding_idx)) outputs = embedding("embedding_tensorIds_paddings", ids, vocab_size, embedding_dim, padding_idx=cpt.long_type(padding_idx), vocab_embeddings=table, compare=False) # print('outputs {}'.format(outputs))