Add homogeneous sampling learner

[jbweston]

(original issue on GitLab)

opened by Anton Akhmerov (@anton-akhmerov) at 2017-07-26T17:17:15.043Z

This should be a very simple prototype that calculates f(x) by sampling the space homogeneously. It isn't adaptive in the sense that its point selection strategy does not depend on the values of f, but it can still be useful for interactive work (the user launches the learner and interrupts it manually when the result looks good).

[jbweston]

originally posted by Anton Akhmerov (@anton-akhmerov) at 2017-07-26T17:54:11.317Z on GitLab

This can be done by making the y-scaling strategy of the 1D learner user-controllable, and allowing to effectively reduce the contribution of y-values to the loss to 0.

[jbweston]

originally posted by Bas Nijholt (@basnijholt) at 2018-02-02T13:45:16.881Z on GitLab

I needed this just now for a Learner1D and it can easily be realized by using:

def loss_per_interval(interval, scale, function_values):
    x_left, x_right = interval
    x_scale, _ = scale
    dx = (x_right - x_left) / x_scale
    return dx

learner = adaptive.Learner1D(f, bounds, loss_per_interval)

[jbweston]

originally posted by Anton Akhmerov (@anton-akhmerov) at 2018-02-02T13:48:31.888Z on GitLab

Ah, cool! I imagine a similar thing would work in 2D, right?

[jbweston]

originally posted by Bas Nijholt (@basnijholt) at 2018-02-02T13:55:12.401Z on GitLab

Yup, this will work:

def loss_per_triangle(ip):
    from adaptive.learner.learner2D import areas
    A = areas(ip)
    return np.sqrt(A)

[jbweston]

originally posted by Anton Akhmerov (@anton-akhmerov) at 2018-02-02T17:45:21.374Z on GitLab

Nice. I'm not sure what's the best way to address the issue then. What about introducing those loss functions in the corresponding modules and documenting their existence?

[jbweston]

originally posted by Bas Nijholt (@basnijholt) at 2018-02-19T16:21:28.553Z on GitLab

This issue is being addressed in https://gitlab.kwant-project.org/qt/adaptive/merge_requests/49.

[jbweston]

originally posted by Joseph Weston (@jbweston) at 2018-02-19T17:54:17.239Z on GitLab

fixed by gitlab:!49