User guide

Here is some information on how to improve your experience with PyConTurb.

Enhancing performance

There are a few ways to reduce the memory impact and increase the speed of your computations.

To reduce memory when calling gen_turb, you could try:

  1. Using np.float32 instead of the default 64-bit float.

  2. Reducing nf_chunk.

  3. Simulating the u, v, and w components separately (if using the exponential coherence model from IEC 61400-1).

To increase speed when calling gen_turb, you could try:

  1. Simulating the u, v, and w components separately (if using the exponential coherence model from IEC 61400-1).

  2. Pre-generating the coherence files and loading them on simulation (if simulating the same geometry multiple times, e.g., with different constraining time series or different random seeds).

  3. Increasing nf_chunk.

Common mistakes

Here is a non-exhaustive list of common issues when simulating with PyConTurb.

  • Simulating component where there is no constraint. Because there is no correlation between the three turbulence components (i.e., u is not correlated to v or w), PyConTurb will default to IEC parameters for the components without a constraint. You can get around this default behaviour by defining a custom sig_func (see examples).

  • Mismatching time steps in constraint and turbulence box. Due to the constrained-turbulence methodology, PyConTurb requires the time steps of the desired simulated turbulence to match the time steps of the constraining time series.

  • Using `interp_data` without understanding interpolator. PyConTurb has the option to attempt to interpolate and of the three profile functions from the constraining time series. This option is provided for convenience, but using it without understanding how the interpolator works can produce unexpected results. In general, I recommend fitting custom profile functions to your data manually instead of using the interpolator.

  • Trying to simulate very large boxes. It is easy to run out of memory if you request a box that is very large. To save memory, you can try setting nf_chunks to 1 (will slow down simulation) or dtype=np.float32.

Bugs and feature requests

This package is a labor of love, and it has by no means achieved perfection. I have written many tests to ensure that the code is acting as expected, but bugs may still be hiding! I am relying on you, the user, to report and/or fix any bugs that you find in the code. Suggestions for improvement are also welcome.

Please do not email me directly for bug reports or feature requests. Instead, use GitLab’s issue tracker. Note that I am unfortunately unable to provide support for issues related to incorrect usage.

Bugs

If PyConTurb is not behaving as you think it should, please submit an issue using GitLab’s issue tracker. NOTE! Please be sure to use the “Bug” template!

Feature requests

If you would like PyConTurb to have a specific feature that it currently doesn’t have, you have two options:

  1. Submit a Feature Request via GitLab’s issue tracker. Be sure to use the “Feature request” template!

  2. Implement the feature in a new branch, and submit a merge request. (See instructions in the Developer corner.)

Because PyConTurb is not under heavy development, priority will be given to merge requests. A new feature request is only likely to be implemented if it is needed in a separate project. Note that if you author a merge request and it is accepted, you will be added to the list of contributors (unless you wish otherwise).