loading pre-trained model

[avieini]

Hi,
I'm trying to load a pretrained model and use it with my own dataset.
How can I do it?

thanks

[MirunaPislar]

Hi,
Which if the models are you trying to load? tf_attention, dnn_models or some other model?

[avieini]

Thanks for the quick answer!
I basically want to run the tf_attention model on my data and visualize the words that are more relevant than other to the discrimination (as in your example ).
my data is a csv file with one columns for the sentence and another binary column indicating if the sentence is sarcastic or not

[MirunaPislar]

Hey. So I've just checked and it should work if you simply clone the repo (you will need all code and the tf_attention model) and just run python3 visualize_tf_attention.py from inside the src/ directory. This will automatically load the tf_attention model specified by MODEL_PATH (by default this is the one that I provided in models/tf_attention).

If you want to use your own data, then place it under /res/tokens. Make sure you rename your files as "tokens_clean_original_train.txt" and "tokens_clean_original_test.txt" and format them in the same way as my files in order to get the visualizations for your data and not mine. After running this, the visualizations will be automatically saved as "attention_vis.html" under the directory plots/html_visualizations/.

More generally, if you want to re-use the model without necessarily using my code, just download the full tf_attention model (link here) and then this would definitely work to restore (any) pre-trained model (you'll need to specify the MODEL_PATH):
`
graph = tf.Graph()

with graph.as_default():

    sess = tf.Session()

    with sess.as_default():

            # Load the saved meta graph and restore variables

            saver = tf.train.import_meta_graph("{}.meta".format(MODEL_PATH))

            sess.run(tf.global_variables_initializer())

            sess.run(tf.local_variables_initializer())

            saver.restore(sess, MODEL_PATH)

            print("Model file restored.")`

Then you can re-adapt my code to get the visualizations you want :) But the first approach would be way faster. Hope this helped!

[avieini]

Hi thank again,
I did as you wrote (the first option) but i'm getting the following error:

Traceback (most recent call last):
File "/home/dave/miniconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1334, in _do_call
return fn(*args)
File "/home/dave/miniconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1319, in _run_fn
options, feed_dict, fetch_list, target_list, run_metadata)
File "/home/dave/miniconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1407, in _call_tf_sessionrun
run_metadata)
tensorflow.python.framework.errors_impl.InvalidArgumentError: Assign requires shapes of both tensors to match. lhs shape= [5382,100] rhs shape= [42912,100]
[[{{node save/Assign_11}} = Assign[T=DT_FLOAT, _class=["loc:@Embedding_layer/Variable"], use_locking=true, validate_shape=true, _device="/job:localhost/replica:0/task:0/device:CPU:0"](Embedding_layer/Variable/Adam_1, save/RestoreV2:11)]]

apparently, something went wrong with the pretrained model and it's graph and my data.
If you have any idea how to overcome it I would be very glad.