Differences
This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision Next revision | Previous revision Next revision Both sides next revision | ||
|
bfe [2016/03/07 17:55] zashi |
bfe [2016/04/05 19:55] zashi [n-Depth Mean Compare] |
||
|---|---|---|---|
| Line 22: | Line 22: | ||
| Programmatically, the read/write head is equivalent to a pointer and the sheet is simply a contiguous block of RAM. | Programmatically, the read/write head is equivalent to a pointer and the sheet is simply a contiguous block of RAM. | ||
| - | In many ways, BF is ideal to use as a genes. Aside from being Turing-complete, which allows it to do anything that can be accomplished with any other programming language, strings of non-functional code can exist and be bypassed. With a single character (command) mutation that bypassed code path can become active, just like actual genetic code consisting of base-pairs. | + | In many ways, BF is ideal to use as programmatic genes. Aside from being Turing-complete, which allows it to do anything that can be accomplished with any other programming language, strings of non-functional code can exist and be bypassed. With a single character (command) mutation that bypassed code path can become active, just like actual genetic code consisting of base-pairs. |
| ====== Specific Design Needs ====== | ====== Specific Design Needs ====== | ||
| Line 28: | Line 28: | ||
| ===== Framework ===== | ===== Framework ===== | ||
| - | BFE is the framework in it's entirety. The framework (BFE) depends on two executable modules and one Tcl proc. One to test strands and one to evaluate their output. The Tcl proc provides mutations to top-scoring strands. The framework consists of a master script and worker script. The master script mediates access to the database and handles logging. The worker script is run on each computation node, generally one process per available CPU core. | + | BFE is the framework in it's entirety. Previous versions of BFE relied on a master script with a slave running for each core in a cluster. |
| + | |||
| + | The latest incarnation consists a single Tcl script using an interpreter with OpenMPI bindings. A stand-alone executable is used for running the bf code and a C-based Tcl extension provides the 'nmc' command used to score output. | ||
| + | |||
| + | BFE tracks all mutations it generates, but when considering the next generation, it groups by score and picks randomly from the top 2% of this population. In this way, different 'species' are picked rather than the same species of strand overfilling the top ranks. | ||
| + | |||
| ===== Mutation Algorithm ===== | ===== Mutation Algorithm ===== | ||
| Line 44: | Line 50: | ||
| ===== n-Depth Mean Compare ===== | ===== n-Depth Mean Compare ===== | ||
| - | Perhaps the greatest challenge in putting BFE together was determining how to score strands. [[:n-Depth_Mean_Compare|n-Depth Mean Compare]] (NMC) compares one ordered set of bytes to another of potentially differing length. NMC returns a floating point greater than 0. The closer to zero, the more similar the sets are. A score of zero means the sets are identical. | + | Perhaps the greatest challenge in putting BFE together was determining how to score strands. [[:n-Depth_Mean_Compare|n-Depth Mean Compare]] (NMC) compares one ordered set of bytes to another of potentially differing length. NMC returns a floating point greater than or equal to 0. The closer to zero, the more similar the sets are. A score of zero means the sets are identical. |
| In function and practice, NMC is similar to calculating [[https://en.wikipedia.org/wiki/Jaccard_index|Jaccard distance]], but also takes into account how far removed the set of bytes are from one another, not just how many bytes are common between the two. To clarify, NMC does not just calculate how similar one set is to another. The lower the n-Depth Mean Compare score, the fewer steps are needed to transform one set to the other. | In function and practice, NMC is similar to calculating [[https://en.wikipedia.org/wiki/Jaccard_index|Jaccard distance]], but also takes into account how far removed the set of bytes are from one another, not just how many bytes are common between the two. To clarify, NMC does not just calculate how similar one set is to another. The lower the n-Depth Mean Compare score, the fewer steps are needed to transform one set to the other. | ||
