vanderschaarlab / mlforhealthlabpub Goto Github PK

I have downgraded my Tensorflow to v1.15 using Minicoda virtual environment but the error still persists it's showing: 'TensorFlow' has no attribute 'reset_default_graph'.
Please help as I'm unavailable to run the code.

Description

Upon inspecting PATE-GAN, I believe there might be an issue with the teachers-discriminators. Algorithm 1 on page 7 from the paper, states that the teachers are supposed to be classifiers which are meant to be continuously trained over the whole training procedure (lines 9-10 in Algorithm 1). Instead, L203 defines classifiers (Logistic Regressions) which are (re-)fit from scratch at each iteration; this might prevent them from continuously learning.

Hello, I'm trying to run Radial Gan on your example to better understand the algorithm and perhaps use in my paper,
Could you please provide the relevant data? I couldn't find in the drive...

Excuse me, when I use survivalquilts, I get an error message: 'RandomSurvivalForest' object has no attribute 'event_times_', the error location is on line 45 of the class_underlyingmodels. And, could have please provide the example data ( preprocessed METABRIC dataset), thank you

Why the ITE block module's loss doesn't contain the GAN_loss， only have the supervised loss?

Hi, thank you for providing the CartPole example in terms of using the AVRIL algorithm.

I wonder could you also provide the code for testing MIMIC-III data using the AVRIL algorithm? I am curious in what way you generated "inputs,targets,a_dim,s_dim" just like the CartPole example usage.

And also, in what way you merged different data frames and conducted a final version for training. I know the dataset is credential but I am really interested in the coding portion since I already got the dataset by application.

I believe your sharing will be beneficial for many of us! Thank you very much!

I have read the article: Attentive State-Space Modeling of Disease Progression and I think the algorithm is very valuable. But I am still not sure how the author applies this probabilistic model to classification? Is it possible to use Bayes' theorem to classify and select the categories with high probability? I was wondering if the model.get_likelihood function was used to calculate the probability of obtaining the observation sequence for each category separately?
Looking forward to your reply! Thanks!

Description

Upon inspecting PATE-GAN, I believe there is an indexing bug on L190, which means that the first data partition is being fed to every teacher. This naturally causes a privacy issue.

Hi,

I've found the Interpole algorithm quite interesting. So, I've been trying to regenerate the results from the paper. However, for the ADNI experiment, the adni.csv file that is required is missing. Could you please either provide the adni.csv file or the processed data files?

Thanks.

Description

Upon inspecting PATE-GAN, I noticed a couple of potential issues:

• The model expects already processed/scaled data (the processing is done outside of the model, i.e., here), and as a consequence the model doesn’t return synthetic data in the original scale.
• The data bounds (min/max values) are directly extracted from the data in a non-DP way (here), which might lead to privacy vulnerabilities as shown in previous work.

I have a question regarding the implementation of the supervised loss training. This is computed in the code as follows:

G_loss_S = tf.losses.mean_squared_error(H[:,1:,:], H_hat_supervise[:,1:,:]), where H_hat_supervise is computed as:

Should not the supervised loss be computed on the latent representation coming from the generator (at least this is how I understood the paper):

H_hat = supervisor(E_hat, T) ?