| … | … | @@ -10,7 +10,7 @@ pathways, involving genes, proteins and other molecules, are graphical |
| 10 | 10 | networks as well. So are social networks. What these all have in common |
| 11 | 11 | is an underlying set of allowed (or preferred) attachments, moves, |
| 12 | 12 | changes or state transitions, as well as hard and soft constraints on |
| 13 | | what attachements, moves are possible. |
| 13 | what attachments, moves are possible. |
| 14 | 14 | |
| 15 | 15 | The goal of the code in this repo is to generate such networks, given |
| 16 | 16 | a (weighted) collection of constraints and/or transition rules and/or |
| … | … | @@ -21,7 +21,7 @@ being of fundamentally different types. In between the start and end |
| 21 | 21 | is a networked web of interconnected moves, transitions, constraints. |
| 22 | 22 | In essence, this is as a constraint satisfaction solver which allows |
| 23 | 23 | probabilistic weighting and guiding, and allows the general problem |
| 24 | | to be described as a collection of syntactical rules or moves (aka |
| 24 | to be described as a collection of syntactical rules or moves (AKA |
| 25 | 25 | rules of inference, axioms, judgments). |
| 26 | 26 | |
| 27 | 27 | Currently focused on natural language generation; however, the theory |
| … | … | @@ -29,7 +29,7 @@ is generic, and is intended for use on any deduction or induction |
| 29 | 29 | problem. The image above was generated with [this example |
| 30 | 30 | code](./examples/export-to-gml.scm); it shows a random network, with |
| 31 | 31 | all links and nodes being of the same type, and the node arity (number |
| 32 | | of connections) being Zipfian-dsitributed (that's why most nodes have |
| 32 | of connections) being Zipfian-distributed (that's why most nodes have |
| 33 | 33 | only one connection, and only one has six connections). |
| 34 | 34 | |
| 35 | 35 | ## Sheaves, germs and syntax |
| … | … | @@ -311,7 +311,7 @@ design choices: |
| 311 | 311 | The current code implements a breadth-first aggregation algorithm. |
| 312 | 312 | It provides two search modes: a deterministic explorer that will |
| 313 | 313 | exhaustively generate all finite "unambiguous" graphs, and an |
| 314 | | random network explorer, useful for sampling from larger grammers |
| 314 | random network explorer, useful for sampling from larger grammars |
| 315 | 315 | that cannot be fully explored via exhaustive enumeration. It passes |
| 316 | 316 | all unit tests. See |
| 317 | 317 | [opencog/generate/README](opencog/generate/) for algo details. |
| … | … | @@ -323,7 +323,7 @@ to might generate an infinite number of graphs. |
| 323 | 323 | |
| 324 | 324 | The random explorer seems to work "well enough", but is still missing |
| 325 | 325 | desirable features. These will be added as the need arises. Foremost |
| 326 | | are additional tunable paramaters to control the statistical |
| 326 | are additional tunable parameters to control the statistical |
| 327 | 327 | distribution of the generated graphs. |
| 328 | 328 | |
| 329 | 329 | This has not yet been tried on anything but small, toy grammars, and |
| … | … | @@ -332,7 +332,7 @@ for large, complex grammars. |
| 332 | 332 | |
| 333 | 333 | Upcoming plans are to provide constraints to generate planar graphs. |
| 334 | 334 | An extension to force linear order (so as to obtain sequences, e.g. |
| 335 | | word sequences) is also equired. |
| 335 | word sequences) is also required. |
| 336 | 336 | |
| 337 | 337 | Some of the missing features are listed in the github issues list. |
| 338 | 338 | |