I have an existing trained model (specifically tensorflow word2vec https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/udacity/5_word2vec.ipynb). I restore the existing model well enough:
model1 = tf.train.import_meta_graph("models/model.meta")
model1.restore(sess, tf.train.latest_checkpoint("model/"))
But I don't know how to use the newly loaded (and trained) model to make predictions. How do I do predictions with a restored model?
Edit:
model code from the official tensorflow repo https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/tutorials/word2vec/word2vec_basic.py
Based on how you are loading the checkpoint I assume this should be the best way to use it for inference.
Load the placeholders:
input = tf.get_default_graph().get_tensor_by_name("Placeholders/placeholder_name:0")
....
Load the op you use to perform prediction:
prediction = tf.get_default_graph().get_tensor_by_name("SomewhereInsideGraph/prediction_op_name:0")
Create a session, execute the prediction op, and feed data in the placeholders.
sess = tf.Session()
sess.run(prediction, feed_dict={input:input_data})
On the other hand, what I prefer to do is always create have the whole model creation inside a constructor of a class. Then, what I would do is the following:
tf.reset_default_graph()
model = ModelClass()
loader = tf.train.Saver()
sess = tf.Session()
sess.run(tf.global_variables_initializer())
loader.restore(sess, path_to_checkpoint_dir)
Since you want to load the embeddings from a trained word2vec model in another model, you should do something like:
embeddings_new_model = tf.Variable(...,name="embeddings")
embedding_saver = tf.train.Saver({"embeddings_word2vec": embeddings_new_model})
with tf.Session() as sess:
embedding_saver.restore(sess, "word2vec_model_path")
Assuming that the embeddings variable in the word2vec model is named embeddings_word2vec.