EMC-EB on Jupyter

[Francisco Ferraz]

Hi:

I've been trying to run the TADPOLE-SHARE code, more concretely the EMC-EB example after properly installing JupyterLab.

While running the cell for "Train model on data set D1 and test model on longitudinal data set D2" works fine, the cell for "Train model on data set D1 and test model on cross sectional data set D3" throws the following error:

Has this happened to anyone before? 

Thanks in advance.

[Vikram]

Hi Francisco,

Thanks for bringing it to our notice. We are currently working on fixing the error. It will probably take around 3 weeks to solve the error (some of the teammates are on vacation). I will keep you updated.

Vikram

[Esther Bron]

Hi Francisco,

It took a bit more time than expected, but the problem for the emceb algorithm is solved. If you pull the newest versions of tadpole-algorithms and jupyter, you should be able to run them (i.e. https://github.com/tadpole-share/jupyter/blob/master/emceb.ipynb)

Please let us know if you encounter any other problems.

Kind regards,

Esther

 

Op dinsdag 16 juni 2020 om 12:50:15 UTC+2 schreef 88.v...@gmail.com:

[Francisco Ferraz]

Hello Esther,

I just tried doing a full instalation and when running the 'pip install git+https://github.com/tadpole-share/tadpole-algorithms' command I'm getting the following error:

My Python version is 3.7.8, is this my fault? Thanks in advance.

Francisco.

[Esther Bron]

Hi Francisco,

Thanks for trying this out! Since the code is currently under heavy development (we had a hackthon to add novel algorithms two weeks ago), there are still some undiscovered bugs, and this is one of them.

It is not your python version, but one of the other libraries that is being installed in the setup of tadpole-algorithms. The library for the novel model  benchmark_smnsr  (by Robert Ciszek from University of Kuopio, Finland) is not yet automatically pip installable, so that is where it fails. Robert is planning to add smnsr to pypi soon.

I made a new push that does not try to install this package that cannot be installed automatically. Could you give it another try?

In case you would like to try the SMNSR algorithm as well (not fully tested yet), you could install the required package using "pip install git+https://github.com/rciszek/SMNSR".

Hope this helps!

Esther

Op zaterdag 1 augustus 2020 om 11:35:54 UTC+2 schreef franf...@gmail.com:

[Francisco Ferraz]

Hi Esther,

This happens when trying pip install git+https://github.com/tadpole-share/tadpole-algorithms:

As far as I know, mkl-service isn't pip installable, only conda installation is referenced in their GitHub. I have tried however, but it doesn't seem to work.

Is there an alternative way to get TADPOLE-SHARE running via Conda?

Thanks for your time.

Fran.

[Esther Bron]

Hi Fran,

This indeeds needs proper testing and fixing before the extension with the novel algorithms is fully automatically installable. I added the bug (#17) in GitHub.

For now, I would recommend you to clone the tadpole-algorithms repository directly:

1. Clone repository jupyter: https://github.com/tadpole-share/jupyter

This repository contains ipython notebooks for running the algorithms. The strengths is that the code in the notebooks is exactly the same for each model.

git clone https://github.com/tadpole-share/jupyter.git

2. Clone repository tadpole-algorithms: https://github.com/tadpole-share/tadpole-algorithms

This repository contains all algorithms, data reading functions and evaluations functions. 

git clone https://github.com/tadpole-share/tadpole-algorithms.git

3. Make sure that the ipython notebooks from the jupyter folder can find the algorithms by adding the location of tadpole-algorithm to your python path.

For example in Visual Studio Code, this can be done by making an .env-file in workspace with the following content: PYTHONPATH=${PYTHONPATH}\path\to\tadpole-algorithms

Esther

Op dinsdag 4 augustus 2020 om 11:40:01 UTC+2 schreef franf...@gmail.com: