rexvalkering/esme

Evolutionary Scheduling Made Easy: an attempt to solve a common combined clustering and rostering problem with evolutionary algorithms.

esme's Issues

Find a module name that is catchier than 'solutionsolver'.

Update the progressbar view such that it shows distinct phases, not steps

Move the evolutionary algorithm methods to a different file.

Add support for different solving phases

For optimal convergence we need to add support for different solving phases. For example, an iterator with different phases might look like this:

1. Solve the clustering problem
2. Solve the clustering and scheduling problem (alternating)
3. Solve the clustering and scheduling problem combined, with lower mutation probability
4. Solve the scheduling problem with lower mutation probability.

Rename clustering to assignment

Split the functionality from the command line interface.

In final report, scores reported are switched.

Add unit tests.

Export the clustering solution to a CSV file.

Currently the clustering solution is simply printed in the terminal. We need to add an option to specify a problem name (e.g. myteams) and use that name to export two files: myteams_teams.csv and myteams_schedule.csv.

Add parameter to determine iteration profile used

Add BetterCodeHub integration.

Add input file validation

Add support for varying number of timeslots per day

In the upcoming test, the number of timeslots differs per day. We could solve this by setting availability to 0 for all participants, but it would be better to enhance the program such that different timeslots per day are supported.

Split visualisation code from running code so the script can be run with pypy.

Currently matplotlib is blocking this.

Create a profiler for profiling or comparing different solution methods

Add an algorithm for clustering.

The original proposal featured both clustering and scheduling. Earlier trials with clustering did not bear fruit.

A solution using evolutionary algorithm would be similar to the scheduling problem. Calculate the maximum group size m and the number of groups n; create a list with each group number repeated m times; the permutation determines the group assignment. Make two distinct assignments for women and men separately.

Normalize the different parameters (e.g. length, motivation, skill) and determine their weights. Find a suitable function for determining fitness. Add extra checks to enforce minimum group sizes.

Fix examples

Examples may be outdated.

Reduce console output to zero if report parameter is False

Add a score and penalty breakdown.

Integrate the progressbar as part of the iterator

Add multi-core support

The deap module supports multi-processor calculations. We should support that too.

Allow CSV's to be read with different separators.

Add output validation

Rename module to esme

Adjust SolverPhase and SolverProgressionPhase to allow for a maximum allotted time.

Refactor code to Python module

Allow passing of a custom progress reporting function

For the demo we want to be able to call esme using Python Celery, and then report on progress using a custom progress bar. For this, we need to store intermediate results, either in file or in the database.

Create a class or function pattern for reporting intermediate progress.

Add subclass of SolverPhase allowing for specifying progress instead of number of generations.

Monitor the progression of the different scores and create a plot at the end of the report.

Default run without arguments should not return an error.

Support input-data from non-csv formats (e.g. xls, xlsx, ods).

Refactor code to comply with at least an 8/10 in BetterCodeHub.

Add support for running the program using Python celery.

Add simple tests as shellscripts.