Save figures from notebook and plot mean/2std of results column

v0.3.1

updated README, minor refactoring and version bump

Switched to mpi4py

• replaced job_stream with mpi4py.
  Job_stream uses threading under the hood which created problems when using multiprocessing in the experiments. With mpi4py this issue is resolved, however, the process with rank 0 is now only distributing the work and gathering the eperiment results.
• Changes to the --ignore_config/-I option: while this option before only affected the skipping of experiments it now also affects the restarting.
• If restoring last state is not successful, ClusterWork now also tries to restore the second to last state before falling back to restarting at iteration 1.
• Added handling of SIGINTs to exit nicely without trace.
• Added some fanciness to the progress report
• Some changes to PlotWork
• Updated README.md

Changes in job_stream pipeline

Changes in the job_stream pipeline:

• The use of frames has been discarded, since they did non work reliably
• All repetitions of all experiments are emitted as work item from the init function
• A reducer takes care of sorting the results into the appropriate DataFrames

Time of iterations and repetitions is now recorded in the results DataFrame automatically.

PlotWork uses tabs instead of accordeon to display results of multiple experiments/repetitions

Logging, plot_work, ...

Many changes, see commits for details

Deep List/Grid Feature

Support for deep dictionaries in the list/grid feature has beed added

Stable Grid and List Evaluations

Besides minor changes and improvements the grid and list evaluation feature has been fixed.

Restarting works now (hopefully)

fixed bug when restarting: index of DataFrame has to be restored.

added some comments and did minor refacturing.

Restart and Progress

Implemented the restart feature and the progress report feature.
The restart feature let's you restart the experiments from each iteration on or skip entire experiments if they have already finished.
The progress report feature loads the log files and evaluates how much of the iterations have already been completed. From that it compiles a progress report and produces an output to the terminal.

Hotfix in Writing Results

Fixed bug that prevented the results to be written in the results file.

First Release

This is the first release of ClusterWork, a framework for constructing and running experiments with different sets of parameters on an MPI-based computing cluster.
The software is based on the Python Experiment Suite by Thomas Rückstieß which can be found here. The relicts of the Python Experiment Suite are that the framework is used by sub-classing and abstract base class. The configuration and the execution of the experiment, however, have changed completely. For the execution on an MPI-based cluster we use the job-stream library which is available via PyPI and github.