Cannot predict properly

[Running-z]

Hello, I updated your latest code, and then used the davis dataset you provided to build two scripts for training and forecasting. The training can be done normally, but I predict that I got an unexpected error. Here is the details. the process of

This is the script code I trained:

CUDA_VISIBLE_DEVICES=2
spec='python3 driver.py --dataset davis \
 --model graphconvreg \
--prot_desc_path davis_data/prot_desc.csv --arithmetic_mean \
--model_dir ./model_dir_tmp3 --plot --aggregate toxcast --csv_out ./outs \
--intermediate_file intermediate_cv2.csv '
eval $spec

Then I use the following script code to make predictions:

CUDA_VISIBLE_DEVICES=2
spec='python3 driver.py --dataset davis --prot_desc_path davis_data/prot_desc.csv \
--model graphconvreg \
--model_dir ./model_dir_tmp3 --predict_only  --restore_model  \
--csv_out ./preds_all_tc_graphconv.csv '
eval $spec

But I predict that I get the following error:

Traceback (most recent call last):
  File "driver.py", line 718, in <module>
    tf.app.run(main=run_analysis, argv=[sys.argv[0]] + unparsed)
  File "/home/zh/anaconda3/envs/deep2.0.0/lib/python3.6/site-packages/tensorflow/python/platform/app.py", line 124, in run
    _sys.exit(main(argv))
  File "driver.py", line 281, in run_analysis
    prediction_file=csv_out)
  File "/project/git3/PADME2/dcCustom/molnet/run_benchmark_models.py", line 191, in model_regression
    restore_model=restore_model)
  File "/project/git3/PADME2/dcCustom/models/tensorgraph/graph_models.py", line 84, in __init__
    super(WeaveModel, self).__init__(**kwargs)
  File "/project/git3/PADME2/dcCustom/models/tensorgraph/tensor_graph.py", line 100, in __init__
    super(TensorGraph, self).__init__(**kwargs)
  File "/project/git3/PADME2/dcCustom/models/models.py", line 59, in __init__
    assert os.path.exists(model_dir)
AssertionError

But in fact, this ./model_dir_tmp3 folder exists, that is, my custom --model_dir is not used.

Then I renamed the model_dir_tmp3 folder to model_dir and executed the following prediction script:

CUDA_VISIBLE_DEVICES=2
spec='python3 driver.py --dataset davis --prot_desc_path davis_data/prot_desc.csv \
--model graphconvreg \
--model_dir ./model_dir --predict_only  --restore_model  \
--csv_out ./preds_all_tc_graphconv.csv '
eval $spec

But I got the following error again:

Traceback (most recent call last):
  File "driver.py", line 717, in <module>
    tf.app.run(main=run_analysis, argv=[sys.argv[0]] + unparsed)
  File "/home/zh/anaconda3/envs/deep2.0.0/lib/python3.6/site-packages/tensorflow/python/platform/app.py", line 124, in run
    _sys.exit(main(argv))
  File "driver.py", line 281, in run_analysis
    prediction_file=csv_out)
  File "/project/git3/PADME2/dcCustom/molnet/run_benchmark_models.py", line 195, in model_regression
    model.predict(train_dataset, transformers=transformers, csv_out=prediction_file, tasks=tasks)
  File "/project/git3/PADME2/dcCustom/models/tensorgraph/tensor_graph.py", line 648, in predict
    self.restore()
  File "/project/git3/PADME2/dcCustom/models/tensorgraph/tensor_graph.py", line 1071, in restore
    saver = tf.train.Saver(var_list=var_list)
  File "/home/zh/anaconda3/envs/deep2.0.0/lib/python3.6/site-packages/tensorflow/python/training/saver.py", line 1239, in __init__
    self.build()
  File "/home/zh/anaconda3/envs/deep2.0.0/lib/python3.6/site-packages/tensorflow/python/training/saver.py", line 1248, in build
    self._build(self._filename, build_save=True, build_restore=True)
  File "/home/zh/anaconda3/envs/deep2.0.0/lib/python3.6/site-packages/tensorflow/python/training/saver.py", line 1272, in _build
    raise ValueError("No variables to save")
ValueError: No variables to save

I am sure that my training is completely normal, and the model training files are saved, like the following

But can you tell me where is wrong?

[simonfqy]

Hi, I apologize that my code is currently a bit dirty. You're welcome to make changes, but if you don't, please do as what I say, and I will try to see whether it is possible to refactor it soon. My workstation needs some set up before I can resume working on it, so it might take some time for me.

For now, if you want to enable --predict_only, you MUST specify its dataset to be nci60 in the bash script. You might need to manually tweak molnet/load_function/nci60_dataset.py so that it suits your need.

Make sure that your working directory is correct. You might need to insert a line pdb.set_trace() inside the models.py script to see what is the directory immediately before the problem occurs.

If you still have any problems after changing it to nci60, please ask me again.