Olivier Verdier
Olivier Verdier
Since start and reset are mutually exclusive, I don't see the need for two different shortcuts for these two actions.
We could override some of numpy's functions such as `sum` or `norm`, to follow the lines of the original chebfun implementation
Currently, only the 2-norm is implemented; all the variation of norms should be implemented, following the syntax of numpy's norm
Many relevant Chebyshev algorithms are now available in https://github.com/numpy/numpy/blob/master/numpy/polynomial/chebyshev.py We should try to use those functions as much as possible.
Right now, the polynomial is converted in a standard basis (1,x,x**2,..) before the roots are computed with a companion matrix. The Chebfun way is to use the Chebyshev basis instead...
The dichotomy algorithm to find the optimal number of Chebyshev coefficients is currently not implemented efficiently. At each dichotomy step, one needs only to compute half the new interpolation values,...
When the approximation is bad (typically, with lots of points), the Chebfun will create a new object, but the underlying barycentric interpolation actually fails. Here is a test: ``` python...
This is more ambitious: functions which are not smooth cannot be properly approximated with chebfuns. The idea is to detect the points where one can split the interval, and then...
Scaling is necessary for Chebfun objects, because the right scaling cannot always be deduced from the data alone. For instance, the following test currently fails: ``` python def runge(x): return...
#### Description :octocat: Most of ranger bindings are overriden by evil #### Reproduction guide :beetle: - Start Emacs - Launch ranger with `SPC j d` - Hit, for instance `RET`...