Implementation on Multi-task BO or BO varying the size of the dataset

[heejincs]

Hi, thank you for the library. This library looks very interesting.

I have read the arvix paper, Fast Bayesian Optimization of Machine Learning Hyperparameters on Large Datasets. I wanted to try the experiments on the paper. The paper says to look for the implementation, Multi-task BO and FABOLAS, here, but I could not find it. Is it not implemented yet? Can you help me to duplicate the experiments? To start with, it seems that the x is all continuous variable whereas we need a discrete variable for the Multi-task BO.

Thank you.

[aaronkl]

Hi sorry the code was only on the development branch but I moved it to the master branch now. Look here for an example how to use Fabolas: https://github.com/automl/RoBO/blob/master/examples/example_fmin_fabolas.py

I will also upload an example for MTBO shortly.

[heejincs]

Thank you for the response.

I am getting

Traceback (most recent call last): File "example_fmin_fabolas.py", line 122, in x_best = fabolas_fmin(objective_function, X_lower, X_upper, num_iterations=100) File "/home/kdrl/hchoi/anaconda2/lib/python2.7/site-packages/robo-0.1-py2.7.egg/robo/fmin.py", line 199, in fabolas_fmin env_kernel = george.kernels.BayesianLinearRegressionKernel(task.n_dims, AttributeError: 'module' object has no attribute 'BayesianLinearRegressionKernel'

I see that george package doesn't have kernels.BayesianLinearRegressionKernel.

Do you know what is the problem? Thank you.

[aaronkl]

Hey, have you installed the dependencies? Fabolas needs our own fork of george and you can install it by: for req in $(cat requirements.txt); do pip install $req; done

[heejincs]

Let me reinstall it. Thanks.

p.s I had a problem with Eigen3. Now it works. Waiting for the MTBO example.

Thank you!

On Jul 12, 2016 12:45 AM, "Aaron Klein" notifications@github.com wrote:

Hey have you installed the dependencies? Fabolas needs our own fork of george and you can install it by for req in $(cat requirements.txt); do pip install $req; done

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/automl/RoBO/issues/33#issuecomment-231963507, or mute the thread https://github.com/notifications/unsubscribe/ATa1zUFihGE0TMhxZS6qvKdCEGEtD5DAks5qU0YKgaJpZM4JIadG .

[heejincs]

Do you know when MTBO example will be uploaded? Sorry for the rushing.

Thank you.

On Tue, Jul 12, 2016 at 1:34 AM, Heejin Choi heejincs@gmail.com wrote:

Let me reinstall it. Thanks. On Jul 12, 2016 12:45 AM, "Aaron Klein" notifications@github.com wrote:

Hey have you installed the dependencies? Fabolas needs our own fork of george and you can install it by for req in $(cat requirements.txt); do pip install $req; done

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/automl/RoBO/issues/33#issuecomment-231963507, or mute the thread https://github.com/notifications/unsubscribe/ATa1zUFihGE0TMhxZS6qvKdCEGEtD5DAks5qU0YKgaJpZM4JIadG .

Sincerely, Heejin Choi

[mks2192]

Hey, have you installed the dependencies? Fabolas needs our own fork of george and you can install it by: for req in $(cat requirements.txt); do pip install $req; done

before installing RoBo i did pip install george is that not a correct way ?

[aaronkl]

the command for req in $(cat requirements.txt); do pip install $req; done should already install the right dependencies. If not could you install the development branch of automl/george