dhsbpp is a tool for solving Dynamical Hierarchical Structured Bin Packing Problem. It supports tuning of various parameters related to algorithm.
go get -u github.com/dati-mipt/dhsbpp
(the same command works for updating)
go run github.com/dati-mipt/dhsbpp/main -max_capacity=100000 -init_days=10 -dataset=australia -separate=root -algorithm=first_fit
Datasets for the algorithm include:
- Tree topology (
datasets/%DATASETNAME%/ChildParent.csv) - Node weights for consecutive epochs (
datasets/%DATASETNAME%/WeightsPerEpoch.csv)
-max_capacity=N, must be specified
Maximum bin size
-init_epochs=N, must be specified
Number of epochs for initial distribution of node weights.
-dataset=<folder>, must be specified
Valid dataset for algorithm
-algorithm=first_fit/greedy, must be specified
Specifies the packing algorithm.
-separate=max_child/root, must be specified
Specifies the way of separating nodes in packing algorithm.
-AF=N, default: 0.6
0 < N < 1
Indicates the maximum allowed usage of a single bin during (re)allocation.
If a single bin has a maximum capacity of MAX, the bin packing algorithms
will be invoked with AF × MAX for the size of a single bin.
Note: AF + RD < 1
-RD=N, default: 0.2
0 < N < 1
Used for reallocations. The trigger for reallocation is the overload
or underload of a bin.
Note: AF + RD < 1
The tool generates images of consecutive tree distributions.
TODO Upload images