mp2893 / rnn_predict Goto Github PK

Hi
I run the example code as follows, and it gives me some errors, could i ask hot to fix it? thanks.
python gru_onehot.py sequences.pkl labels.pkl outfile
Loading data ... done!!
Building the model ... done!!
Constructing the optimizer ... done!!
Optimization start !!
Traceback (most recent call last):
File "gru_onehot.py", line 268, in
train_GRU_RNN(dataFile=dataFile, labelFile=labelFile, outFile=outFile, inputDimSize=inputDimSize, hiddenDimSize=hiddenDimSize, max_epochs=max_epochs, L2_reg=L2_reg, batchSize=batchSize, use_dropout=use_dropout)
File "gru_onehot.py", line 245, in train_GRU_RNN
validAuc = calculate_auc(test_model, validSet)
File "gru_onehot.py", line 182, in calculate_auc
auc = roc_auc_score(list(labels), list(scoreVec))
File "/usr/local/lib/python2.7/dist-packages/sklearn/metrics/ranking.py", line 260, in roc_auc_score
sample_weight=sample_weight)
File "/usr/local/lib/python2.7/dist-packages/sklearn/metrics/base.py", line 75, in _average_binary_score
return binary_metric(y_true, y_score, sample_weight=sample_weight)
File "/usr/local/lib/python2.7/dist-packages/sklearn/metrics/ranking.py", line 251, in _binary_roc_auc_score
raise ValueError("Only one class present in y_true. ROC AUC score "
ValueError: Only one class present in y_true. ROC AUC score is not defined in that case.

When I execute gru_onehot.py, I get a below error.

File "gru_onehot.py", line 211
print 'Loading data ... ',
^
SyntaxError: Missing parentheses in call to 'print'

Hello Ed,

Our team is working on EMR/EHR data and then applying Machine learning to build models to classify/predict for a specific use cases.
We have gone through your various works such as MedGAN, Medical concept vector using skip-gram and Med2Vec and are really interested in leveraging the same in our experiments.

There are two challenges that we face -

1. Capturing the temporality of data.
2. Deal with large number of features (Diagnosis , medications, Procedures, observations etc.)

We understand that Med2Vec is a better approach than MCV(Skip-gram) as Med2Vec captures the temporality by capturing visit-level representation.

We wanted to understand more on the Medical concept vector , how we can generate medical concept vectors for any EHR data.
Is MCV with skipgram code is already available on github?As we understand that you are using emb.pkl pickle files in your python code .

Thanks,
Ankit