New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ReadTheDocs is failing to build #26
Comments
Just to clarify, the problem is that the name It sounds like Do you know why it's failing in some environments but not on our Travis builds (which check 2.7 and 3.5)? |
Oh hey check this out and, even better, the doc page it links to. |
I think this could be a good solution! I'm going to try it. |
Is it feasible for you to try the builtins-when-setup-is-not-main branch of autograd? I applied the fix linked above and I think it might solve your build problems. |
I'll give it a shot. |
I just installed that branch into a clean environment with all of pycalphad's other dependencies and I'm going to play with the requirements.txt file on RTD and see if I can't bypass this issue by preinstalling autograd. |
I'm pretty confident that new setup.py will fix the error you were running into, and it didn't break anything in autograd, so I just merged it into the master branch. |
Happy to help! Cool project! |
Previously we would error when trying to install and numpy was not installed. This now just goes ahead and installs numpy for you before building the extension.
* ENH: Support restarting from a previous chain A new calculation can be restarted from a previous chain. As a side effect, the chain we output to file is the real chain and not the flat chain. * ENH: Raise if tracefile/probfile will be overwritten Because MCMC fitting is relatively long-running and RAM intensive, it is ideal to be able to restart from a previous calculation. Currently, to restart, a user would manually update the parameters in the database to the parameters with the lowest probability from the previous run, then use that set of parameters to initialize a new set of walkers. Since we save the state of the chains as the tracefile anyways, we can use the chains to restart a run in emcee and avoid tuning the walker initialization parameters on restart. To make this (and analysis) easier, chains and log probabilities are now written out as binaries, rather than text files. This preserves the shape with the number of chains, so the chain file is now of shape (k, n, p), where k is number of walkers, n is the preallocated number of iterations, p is the number of parameters. The previous shape was (kn, p). The same is true for the log probability file, which is now of shape (k, n) rather than (kn). This change also introduces a command line argument --restart which takes the path to a chain file. If the name of the input chain conflicts with the default name, an error will be raised.
https://readthedocs.org/projects/pycalphad/builds/3465916/
The text was updated successfully, but these errors were encountered: