@@ -1,6 +1,6 @@ from __future__ import division import networkx as nx, numpy as np, tempfile -from string import ascii_letters +from string import ascii_letters,digits from collections import namedtuple ############################## @@ -12,8 +12,8 @@ Single # for descriptions of what I'm doing inside source code, and what variabl SYMBOL_LIST = [] #list of pairs (char, bounds) where bounds = (xmin, xmax, ymin, ymax) -HUGE = 9999 -SUPERHUGE = 999999999 +HUGE = 1e10 +SUPERHUGE = 1e20 Symbol = namedtuple("Symbol","char bounds") Parse = namedtuple("Parse","tree score") @@ -31,12 +31,7 @@ class memoized: else: object = self.cache[args] = self.fn(*args) return object - - def __repr__(self): - """Return the function's docstring.""" - return self.fn.__doc__ - - + def parse_score(rule,bounds1, nonterminal1, bounds2, nonterminal2): """computes score just from that particular parse""" @@ -44,17 +39,14 @@ def parse_score(rule,bounds1, nonterminal1, bounds2, nonterminal2): return np.dot(features(rule,bounds1,bounds2),weights(rule)) def features(rule,bounds1,bounds2): + "return list of feature functions for rule" xmin1,xmax1,ymin1,ymax1 = bounds1 xmin2,xmax2,ymin2,ymax2 = bounds2 xmin,xmax,ymin,ymax = bounding_box([bounds1,bounds2]) return [feature(xmin,xmax,ymin,ymax,xmin1,xmax1,ymin1,ymax1,xmin2,xmax2,ymin2,ymax2) for feature in FEATURES[rule]] -CHARACTERS = [char for char in ascii_letters] + [str(i) for i in range(10)] + ["+","-","="] - - - - +CHARACTERS = list(ascii_letters) + list(digits) + ["+","-","="] def test_parse(): @@ -76,15 +68,16 @@ def test_parse(): def parse_expr(char_bound_list): + "top level parse function, applied to list of pairs (char, bound_box)" global SYMBOL_LIST SYMBOL_LIST = [Symbol(char,bounds) for (char,bounds) in char_bound_list] return subset2parse("Expr",tuple(range(len(char_bound_list)))) @memoized def subset2parse(nonterminal,symbol_ind_list): + "recursive parsing function" best_rule, best_tree, best_score = None, None, -HUGE - if len(symbol_ind_list) == 1: return SYMBOL_LIST[symbol_ind_list[0]].char,0 @@ -96,10 +89,6 @@ def subset2parse(nonterminal,symbol_ind_list): tree2,score2 = subset2parse(nt2,comp) cum_score = (score1 + score2 + parse_score(rule,subset_bounds(subset),nt1,subset_bounds(comp),nt2)) - print rule,nt1,nt2 - print cum_score - print symbol_ind_list - if cum_score > best_score: best_rule = rule @@ -110,6 +99,7 @@ def subset2parse(nonterminal,symbol_ind_list): return Parse(best_tree,best_score) def tex_combine(rule,tree1,tree2): + "use appropriate latex syntax to combine subexpressions" return TEX_OPS[rule](tree1,tree2) TEX_OPS = dict( @@ -147,6 +137,8 @@ SupFeats = [LL1,BB1,TT1,R1L2,TT2,Bigger12,RR2] SubFeats = [LL1,BB1,TT1,R1L2,Bigger12,BB2,RR2] +# Each feature function is generated from the following template, which has a fixed set of inputs +# %s is replaced by a string expression that evaluates to a boolean feat_template = """ def _feat(xmin,xmax,ymin,ymax,xmin1,xmax1,ymin1,ymax1,xmin2,xmax2,ymin2,ymax2): height = ymax-ymin @@ -155,13 +147,13 @@ def _feat(xmin,xmax,ymin,ymax,xmin1,xmax1,ymin1,ymax1,xmin2,xmax2,ymin2,ymax2): return %s """ - - def str2feat(math_expr): + "Makes feature function from string expression" exec(feat_template%math_expr) return _feat def makefeats(li): + "Turns list of feature strings into list of feature functions" return map(str2feat,li) FEATURES = dict( @@ -174,18 +166,21 @@ def line2rule(string): pieces = string.split() return Replacement(pieces[0],pieces[2],pieces[3:]) - RULES = [(rep.left,(rep.op,rep.rightlist[0],rep.rightlist[1])) for rep in map(line2rule,PRETTY_GRAMMAR)] @memoized def replacements2(nonterminal): + """Return a list of triples (operator,right1,right2) for every replacement rule with + two items on RHS""" return [(op,right1,right2) for (left,(op,right1,right2)) in RULES if left == nonterminal] - + def subset_bounds(symbol_ind_list): + "Get bounds of a list of symbol indices" return bounding_box([SYMBOL_LIST[i].bounds for i in symbol_ind_list]) @memoized def subsets(tup): + "All subsets of elements of tuples" if tup == (): return ((),) else: later = subsets(tup[1:]) @@ -193,7 +188,8 @@ def subsets(tup): @memoized def subsets_and_comps(tup): - return tuple((ss,tuple(el for el in tup if el not in ss)) for ss in subsets(tup) if increasing(0,len(ss),len(tup))) + "a list of pairs subset,complement for all proper subsets of tup" + return [(ss,tuple(el for el in tup if el not in ss)) for ss in subsets(tup) if increasing(0,len(ss),len(tup))] def increasing(a,b,c): return a<b and b<c layout-parse.py 965e44692dbccd1a11dcf6f5dfe3e0e5e9cba8ed John Schulman schulmannerism@gmail.com http://github.com/joschu/wyoming/commit/6eb4b1eab8310bb9bbb7da0f6777e3f7de9f9d90 6eb4b1eab8310bb9bbb7da0f6777e3f7de9f9d90 2009-11-08T11:31:00-08:00 2009-11-08T11:31:00-08:00 added docstrings 8e34c13999bccc459b97e6bb167bbc8e5cd54564 John Schulman schulmannerism@gmail.com