Changeset 1229
- Timestamp:
- 04/30/23 02:11:46 (12 months ago)
- Location:
- cpp/frams/genetics/f4
- Files:
-
- 3 edited
-
f4_general.cpp (modified) (12 diffs)
-
f4_general.h (modified) (2 diffs)
-
f4_oper.cpp (modified) (3 diffs)
Legend:
- Unmodified
- Added
- Removed
-
cpp/frams/genetics/f4/f4_general.cpp
r1228 r1229 25 25 void rolling_dec(double *v) 26 26 { 27 *v -= 0.7853; // 0.7853981 45 degrees 27 *v -= 0.7853; // 0.7853981 45 degrees = pi/4 like in f1 28 28 } 29 29 … … 40 40 { 41 41 if (i >= slen) // ran out the string, should never happen with a correct string 42 return 1; 42 return 1; //TODO MacKo 2023-04: interesting: why was this situation made undistinguishable from s[1]==stopchar ? does this have any bad consequences or is "1" just used to tell "advance as little as possible"? Anyway, this function can be eliminated when parsing is simplified. 43 43 if (stopchar == s[i]) // bumped into stopchar 44 44 return int(i); 45 45 if (i < slen - 1) // s[i] is not the last char 46 46 { 47 if (s[i] == '(') 47 if (s[i] == '(') //not an allowed char in f4, perhaps a remnant of old experiments with code 48 48 { 49 49 i += 2 + scanRecur(s + i + 1, slen - i - 1, ')'); … … 192 192 // the current genotype code is processed 193 193 //genoRange.add(gcur->pos,gcur->pos+gcur->name.length()-1); 194 bool neuclasshandler = false; // if set to true, then a separate neuron handler below will identify the neuroclass and assign the cell to the neuron type195 194 196 195 // To detect what genes are valid neuroclass names, but do NOT have is_neuroclass==true 197 // (just as a curiosity to ensure we properly distinguish between, for example, the "G" neuron and "G" modifier):196 // (just as a curiosity to ensure we properly distinguish between, for example, the "G" neuron and the "G" modifier): 198 197 //char *TMP = (char*)gcur->name.c_str(); 199 //if (gcur->is_neuroclass==false && GenoOperators::parseNeuroClass(TMP, ModelEnum::SHAPETYPE_BALL_AND_STICK))198 //if (gcur->is_neuroclass==false && GenoOperators::parseNeuroClass(TMP, ModelEnum::SHAPETYPE_BALL_AND_STICK)) 200 199 // printf("Could be a valid neuroclass, but is_neuroclass==false: %s\n", gcur->name.c_str()); 201 200 202 if (gcur->name.length() == 1 && gcur->neuclass == NULL) //one-character genes and not neuroclass names 203 { 201 if (gcur->neuclass == NULL) //not a neuron 202 { 203 if (gcur->name.length() > 1) 204 logPrintf("f4_Cell", "oneStep", LOG_WARN, "Multiple-character code that is not a neuron class name: '%s'", gcur->name.c_str()); //let's see an example of such a code... 205 204 206 genoRange.add(gcur->pos, gcur->pos); 205 207 char name = gcur->name[0]; … … 542 544 default: 543 545 { 544 // because there are one-character neuron classes, default is move control to neuclasshandler545 neuclasshandler = true;546 }547 }548 }549 else550 {551 // if many characters or single character but is_neuroclass, then it will be handled below552 neuclasshandler = true;553 }554 555 if (neuclasshandler)556 {557 genoRange.add(gcur->pos, gcur->pos + int(gcur->name.length()) + 2 - 1); // +2 for N:558 if (type != CELL_UNDIFF)559 {560 // fix: delete this node561 org->setRepairRemove(gcur->pos, gcur);562 return 1; // stop563 }564 // error: if no previous565 if (dadlink == NULL)566 {567 // fix: delete it568 org->setRepairRemove(gcur->pos, gcur);569 return 1; // stop570 }571 // multiple characters are neuron types. Let's check if exists in the current configuration of Framsticks572 char *temp = (char*)gcur->name.c_str();573 neuclass = GenoOperators::parseNeuroClass(temp, ModelEnum::SHAPETYPE_BALL_AND_STICK);574 if (neuclass == NULL)575 {576 546 // error: unknown code 577 547 string buf = "Unknown code '" + gcur->name + "'"; … … 580 550 return 1; 581 551 } 552 } 553 } 554 else 555 { 556 genoRange.add(gcur->pos, gcur->pos + int(gcur->name.length()) + 2 - 1); // +2 for N: 557 if (type != CELL_UNDIFF) 558 { 559 // fix: delete this node 560 org->setRepairRemove(gcur->pos, gcur); 561 return 1; // stop 562 } 563 // error: if no previous 564 if (dadlink == NULL) 565 { 566 // fix: delete it 567 org->setRepairRemove(gcur->pos, gcur); 568 return 1; // stop 569 } 570 neuclass = gcur->neuclass; 582 571 type = CELL_NEURON; 583 572 // change of type also halts development, to give other … … 1300 1289 len = out.length(); 1301 1290 if (len > 1) 1302 if (out[len - 1] == '>') { (out.directWrite())[len - 1] = 0; out.endWrite(); }; 1291 if (out[len - 1] == '>') { (out.directWrite())[len - 1] = 0; out.endWrite(); }; //Macko 2023-04 TODO "can be omitted", but should we remove it as a rule even in generated genotypes? see if I can somehow detect junk characters after top-level '>' ends properly: /*4*/<X>N:N>whatever 1303 1292 // copy back to string 1304 1293 // if new is longer, reallocate buf … … 1340 1329 // scan genotype string and build tree 1341 1330 // return >1 for error (errorpos) 1342 int f4_processRecur(const char* genot, unsigned pos0, f4_Node *parent)1343 { 1344 unsigned int gpos ;1345 f4_Node *par ;1346 1347 gpos = pos0;1348 par = parent;1349 if (gpos >= strlen(genot)) return 1; 1331 int f4_processRecur(const char* genot, unsigned int pos0, f4_Node *parent) 1332 { 1333 unsigned int gpos = pos0; //MacKo 2023-04 (TODO): these two variables are often updated before return which has no effect since they are local. Seems like a half step towards making them (or just gpos) in/out parameter which would solve many issues and simplify parsing (getting rid of scanRecur()) while making it more strict. 1334 f4_Node *par = parent; 1335 1336 if (gpos >= strlen(genot)) 1337 return (int)strlen(genot) + 1; 1338 1350 1339 while (gpos < strlen(genot)) 1351 1340 { … … 1369 1358 else // ran out 1370 1359 { 1371 node = new f4_Node(">", par, int(strlen(genot)) - 1); 1372 par = node; 1360 //MacKo 2023-04, more strict behavior: instead of silent repair (no visible effect to the user, genotype stays invalid but is interpreted and reported as valid), we now point out where the error is. For example <X> or <X><X or <X><N:N> 1361 return gpos + 1; //the problem starts here, occurs because second child (branch) <1..>2..> is not completed 1362 //old silent repair: 1363 //node = new f4_Node(">", par, int(strlen(genot)) - 1); 1364 //par = node; 1373 1365 } 1374 1366 gpos++; … … 1379 1371 f4_Node *node = new f4_Node(">", par, gpos); 1380 1372 par = node; 1381 gpos = (unsigned int)strlen(genot);1373 //gpos = (unsigned int)strlen(genot); //MacKo 2023-04: first of all, 'gpos' is a local variable so no effect; second, '>' may be internal (i.e., not the last one in the genotype), so it is a bad hint to assign strlen(). 'par' above is also local... 1382 1374 return 0; // OK 1383 1375 } … … 1394 1386 gpos += end - (genot + gpos); 1395 1387 //gpos++; 1396 //while ((genot[gpos] >= '0') && (genot[gpos] <= '9')) gpos++; node1 = new f4_Node("#", par, oldpos);1388 //while ((genot[gpos] >= '0') && (genot[gpos] <= '9')) gpos++; node1 = new f4_Node("#", par, oldpos); 1397 1389 f4_Node *node = new f4_Node("#", par, oldpos); 1398 1390 node->reps = reps; … … 1407 1399 else // ran out 1408 1400 { 1409 node = new f4_Node(">", par, int(strlen(genot)) - 1); 1401 return gpos; //MacKo 2023-04: report an error, better to be more strict instead of a silent repair (genotype stays invalid but is interpreted and reported as valid) with non-obvious consequences? 1402 //earlier apporach - silently treating this problem (we don't ever see where the error is because it gets corrected in some way here, while parsing the genotype, and error location in the genotype is never reported): 1403 //node = new f4_Node(">", par, int(strlen(genot)) - 1); // check if needed and if this is really the best repair operation; seemed to happen too many times in succession for some genotypes even though they were only a result of f4 operators, not manually created... and the operators should not generate invalid genotypes, right? Or maybe crossover does? Seems like too many #N's for closing >'s; removing #N or adding > helped. Operators somehow don't do it properly sometimes? But F4_ADD_REP adds '>'... (TODO) 1410 1404 } 1411 1405 return 0; // OK … … 1491 1485 1492 1486 // should end with a '>' 1493 if (par) 1494 { 1495 if (par->name != ">") 1496 { 1497 f4_Node *node = new f4_Node('>', par, int(strlen(genot)) - 1); 1498 par = node; 1499 } 1487 if (par && par->name != ">") 1488 { 1489 //happens when gpos == strlen(genot) 1490 //return gpos; //MacKo 2023-04: could report an error instead of silent repair, but repair operators only work in Cells (i.e., after the f4_Node tree has been parsed without errors and Cells can start developing) so we don't want to make a fatal error because of missing '>' here. Also after conversions from Cells to text, trailing '>' is deliberately removed... and also the simplest genotype is officially X, not X>. 1491 f4_Node *node = new f4_Node('>', par, int(strlen(genot)) - 1); 1492 par = node; 1500 1493 } 1501 1494 -
cpp/frams/genetics/f4/f4_general.h
r1228 r1229 42 42 43 43 /** 44 * TODO MacKo 2023-04: not sure if this function is needed and if f4_processRecur() would not suffice 45 * if it advanced the string pointer (in/out parameter) while processing. Its returned value is always used after 46 * f4_processRecur() anyway, and in two cases likely incorrectly (for [...] to detect closing ']' 47 * and for :...: to detect closing ':') - we don't need recursion in these cases, a simple linear 48 * scan would suffice, but even this would not be needed - since we are parsing the actual characters in these cases, 49 * we do scanning anyway. So looks like this function doubles the work already done more thoroughly by f4_processRecur(). 50 * 44 51 * Scans f4 genotype string for a stopping character and returns the position of 45 52 * this stopping character or 1 if the end of string was reached. This method is used … … 494 501 * @return 0 if processing was successful, otherwise returns the position of an error in the genotype 495 502 */ 496 int f4_processRecur(const char *genot, unsigned pos0, f4_Node *parent);503 int f4_processRecur(const char *genot, unsigned int pos0, f4_Node *parent); 497 504 498 505 /** -
cpp/frams/genetics/f4/f4_oper.cpp
r1228 r1229 16 16 // 17 17 // TODO the behavior of neuron input indexes during mutation seems badly implemented (see also TREAT_BAD_CONNECTIONS_AS_INVALID_GENO). Are they kept properly maintained when nodes are added and removed? This could be done well because during mutation we operate on the tree structure with cross-references between nodes (so they should not be affected by local changes in the tree), and then convert the tree back to string. Yet, the f4_Node.conn_from is an integer and these fields in nodes do not seem to be maintained on tree node adding/removal... change these integer offsets to references to node objects? But actually, do the offsets that constitute relative connection references concern the f4_Node tree structure (and all these sophisticated calculations of offsets during mutation are useful) or rather they concern the f4_Cells development? verify all situations in f4_Cell::oneStep(), case '['. 18 // TODO add simplifying sequences of modifiers (so capital and small letter cancel out, like in f1) - but seems like each single modifier is a separate f4_Node? and perhaps we don't want to use the repair mechanism for this... maybe mutations, when they add/modify/remove a modifier node, should be "cleaning" the tree by removing nodes when they encounter contradictory modifiers on the same subpath ?18 // TODO add simplifying sequences of modifiers (so capital and small letter cancel out, like in f1) - but seems like each single modifier is a separate f4_Node? and perhaps we don't want to use the repair mechanism for this... maybe mutations, when they add/modify/remove a modifier node, should be "cleaning" the tree by removing nodes when they encounter contradictory modifiers on the same subpath, and also limit the number of modifiers of each type just like in f1? To avoid squences like ...<X>llmlIilImmimiimmimifmfl<fifmmimilimmmiimiliffmfliIfififlliflimfliffififmiffmfliflifmIlimimiflimfiffmllliflmimifllifliliflifmIlimimiflimfiffmllliflmimifllfmIlimimiflimfiffmllliflmimiflliflimimmiflimfliffmiflifmfiffllIlififliffififmiffmfliflifIliflimimflimflfflimimifllfflifllfflimlififfiiffifIr<r<... 19 19 // TODO add support for properties of (any class of) neurons - not just sigmoid/force/intertia (':' syntax) for N 20 20 // TODO add mapping genotype character ranges for neural [connections] … … 120 120 int Geno_f4::validate(char *& geno, const char *genoname) 121 121 { 122 // convert geno to tree, then try to validate 20 times122 // convert geno to a tree, then try to validate 123 123 f4_Node root; 124 124 if (f4_processRecur(geno, 0, &root) || root.childCount() != 1) return GENOPER_OK; // cannot repair 125 if (ValidateRec(&root, 20) == GENOPER_REPAIR) // if repaired, make it back to string 125 126 const int VALIDATE_TRIALS = 20; 127 if (ValidateRec(&root, VALIDATE_TRIALS) == GENOPER_REPAIR) // if repaired, make it back to string 126 128 { 127 129 geno[0] = 0; … … 674 676 if (f4_processRecur(g2, 0, &root2) || (root2.childCount() != 1)) return GENOPER_OPFAIL; 675 677 676 // decide amounts of crossover, 0.25-0.75 677 // adam: seems 0.1-0.9 -- MacKo 678 // decide amounts of crossover, 0.1-0.9 678 679 chg1 = 0.1 + rndDouble(0.8); 679 680 chg2 = 0.1 + rndDouble(0.8);
Note: See TracChangeset
for help on using the changeset viewer.
