-
Notifications
You must be signed in to change notification settings - Fork 2
/
sys1dev.py
380 lines (295 loc) · 13.5 KB
/
sys1dev.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
#!/usr/bin/env python3
"""
Tweaked code for Arizona submission to sigmorphon2023part1
Alice Kwak, Mike Hammond, Cheyenne Wing
***
Non-neural baseline system for the SIGMORPHON 2022 Shared Task 0.
Author: Mans Hulden
Modified by: Tiago Pimentel
Modified by: Jordan Kodner
Last Update: 22/04/2022
"""
import sys, os, getopt, re
from functools import wraps
from glob import glob
def hamming(s,t):
"""number of differences between two strings of equal length (mh)"""
return sum(1 for x,y in zip(s,t) if x != y)
def halign(s,t):
"""Align two strings by Hamming distance."""
slen = len(s)
tlen = len(t)
minscore = len(s) + len(t) + 1
for upad in range(0, len(t)+1):
upper = '_' * upad + s + (len(t) - upad) * '_'
lower = len(s) * '_' + t
score = hamming(upper, lower)
if score < minscore:
bu = upper
bl = lower
minscore = score
for lpad in range(0, len(s)+1):
upper = len(t) * '_' + s
lower = (len(s) - lpad) * '_' + t + '_' * lpad
score = hamming(upper, lower)
if score < minscore:
bu = upper
bl = lower
minscore = score
zipped = zip(bu,bl)
newin = ''.join(i for i,o in zipped if i != '_' or o != '_')
newout = ''.join(o for i,o in zipped if i != '_' or o != '_')
return newin, newout
def levenshtein(s, t, inscost = 1.0, delcost = 1.0, substcost = 1.0):
"""Recursive implementation of Levenshtein, with alignments returned."""
@memolrec
def lrec(spast, tpast, srem, trem, cost):
if len(srem) == 0:
return spast + len(trem) * '_', tpast + trem, '', '', cost + len(trem)
if len(trem) == 0:
return spast + srem, tpast + len(srem) * '_', '', '', cost + len(srem)
addcost = 0
if srem[0] != trem[0]:
addcost = substcost
return min((lrec(spast + srem[0], tpast + trem[0], srem[1:], trem[1:], cost + addcost),
lrec(spast + '_', tpast + trem[0], srem, trem[1:], cost + inscost),
lrec(spast + srem[0], tpast + '_', srem[1:], trem, cost + delcost)),
key = lambda x: x[4])
answer = lrec('', '', s, t, 0)
return answer[0],answer[1],answer[4]
def memolrec(func):
"""Memoizer for Levenshtein."""
cache = {}
@wraps(func)
def wrap(sp, tp, sr, tr, cost):
if (sr,tr) not in cache:
res = func(sp, tp, sr, tr, cost)
cache[(sr,tr)] = (res[0][len(sp):], res[1][len(tp):], res[4] - cost)
return sp + cache[(sr,tr)][0], tp + cache[(sr,tr)][1], '', '', cost + cache[(sr,tr)][2]
return wrap
def alignprs(lemma, form):
"""Break lemma/form into three parts:
IN: 1 | 2 | 3
OUT: 4 | 5 | 6
1/4 are assumed to be prefixes, 2/5 the stem, and 3/6 a suffix.
1/4 and 3/6 may be empty.
"""
al = levenshtein(lemma, form, substcost = 1.1) # Force preference of 0:x or x:0 by 1.1 cost
alemma, aform = al[0], al[1]
# leading spaces
lspace = max(len(alemma) - len(alemma.lstrip('_')), len(aform) - len(aform.lstrip('_')))
# trailing spaces
tspace = max(len(alemma[::-1]) - len(alemma[::-1].lstrip('_')), len(aform[::-1]) - len(aform[::-1].lstrip('_')))
return alemma[0:lspace], alemma[lspace:len(alemma)-tspace], alemma[len(alemma)-tspace:], aform[0:lspace], aform[lspace:len(alemma)-tspace], aform[len(alemma)-tspace:]
def prefix_suffix_rules_get(lemma, form):
"""Extract a number of suffix-change and prefix-change rules
based on a given example lemma+inflected form."""
lp,lr,ls,fp,fr,fs = alignprs(lemma, form) # Get six parts, three for in three for out
# Suffix rules
ins = lr + ls + ">"
outs = fr + fs + ">"
srules = set()
for i in range(min(len(ins), len(outs))):
srules.add((ins[i:], outs[i:]))
srules = {(x[0].replace('_',''), x[1].replace('_','')) for x in srules}
# Prefix rules
prules = set()
if len(lp) >= 0 or len(fp) >= 0:
inp = "<" + lp
outp = "<" + fp
for i in range(0,len(fr)):
prules.add((inp + fr[:i],outp + fr[:i]))
prules = {(x[0].replace('_',''), x[1].replace('_','')) for x in prules}
return prules, srules
def apply_best_rule(lemma, msd, allprules, allsrules):
"""Applies the longest-matching suffix-changing rule given an input
form and the MSD. Length ties in suffix rules are broken by frequency.
For prefix-changing rules, only the most frequent rule is chosen."""
bestrulelen = 0
base = "<" + lemma + ">"
if msd not in allprules and msd not in allsrules:
return lemma # Haven't seen this inflection, so bail out
if msd in allsrules:
applicablerules = [(x[0],x[1],y) for x,y in allsrules[msd].items() if x[0] in base]
if applicablerules:
bestrule = max(applicablerules, key = lambda x: (len(x[0]), x[2], len(x[1])))
base = base.replace(bestrule[0], bestrule[1])
if msd in allprules:
applicablerules = [(x[0],x[1],y) for x,y in allprules[msd].items() if x[0] in base]
if applicablerules:
bestrule = max(applicablerules, key = lambda x: (x[2]))
base = base.replace(bestrule[0], bestrule[1])
base = base.replace('<', '')
base = base.replace('>', '')
return base
def numleadingsyms(s, symbol):
return len(s) - len(s.lstrip(symbol))
def numtrailingsyms(s, symbol):
return len(s) - len(s.rstrip(symbol))
###############################################################################
def main(argv):
#command-line options, mh
options,remainder = getopt.gnu_getopt(
argv[1:],
'othp:',
['output','test','help','path=']
)
#set a reasonable default path
#TEST,OUTPUT,HELP,path = False,False,False,'../../part1/development_languages/'
TEST,OUTPUT,HELP = False,False,False
#path = '/home/hammond/Desktop/2023InflectionST-main/part1/data/'
path = '/Users/hammond/Desktop/2023InflectionST-main/part1/data/'
for opt, arg in options:
if opt in ('-o', '--output'):
OUTPUT = True
if opt in ('-t', '--test'):
TEST = True
elif opt in ('-h', '--help'):
HELP = True
elif opt in ('-p', '--path'):
path = arg
if HELP:
print("\n*** Baseline for the SIGMORPHON 2022 shared task ***\n")
print("By default, the program runs all languages only evaluating accuracy.")
print("To create output files, use -o")
print("The training and dev-data are assumed to live in ./part1/development_languages/")
print("Options:")
print(" -t evaluate on test instead of dev")
print(" -o create output files with guesses (and don't just evaluate)")
print(" -p [path] data files path. Default is ./part1/development_languages/")
quit()
totalavg,totalavg_seenlemma,totalavg_seenmsd,totalavg_seenneither = 0,0,0,0
numlang,numlang_seenlemma,numlang_seenmsd,numlang_seenneither = 0,0,0,0
mhlangs = sorted(list( #mh
{re.sub('\.trn.*$','',d) for d in os.listdir(path) if '.trn' in d} #mh
)) #mh
print("Lang:\tAll\tLemmas\tMSD\tNeither")
for lang in sorted(
#list({re.sub('\.trn.*$','',d) for d in os.listdir(path) if '.trn' in d})
mhlangs
):
#prefix rules, suffix rules
allprules, allsrules = {}, {}
#check that there's training data
#if not os.path.isfile(path + lang + ".hall.trn"): #mh: for hallucinated
if not os.path.isfile(path + lang + ".trn"):
continue
#read the file
#lines = [line.strip() for line in open(path + lang + ".hall.trn", "r") if line != '\n'] #mh: for hallucinated
lines = [line.strip() for line in open(path + lang + ".trn", "r") if line != '\n']
trainlemmas = set()
trainmsds = set()
# First, test if language is predominantly suffixing or prefixing
# If prefixing, work with reversed strings
prefbias, suffbias = 0,0
for l in lines:
#different order of fields in 2023
#lemma, form, msd = l.split(u'\t')
lemma, msd, form = l.split(u'\t')
#add lemma to set of lemmas
trainlemmas.add(lemma)
#add msd to set.
trainmsds.add(msd)
aligned = halign(lemma, form)
# _ on left is prefixed; _ on right is suffixed
if ' ' not in aligned[0] and ' ' not in aligned[1] and '-' not in aligned[0] and '-' not in aligned[1]:
prefbias += numleadingsyms(aligned[0],'_') + numleadingsyms(aligned[1],'_')
suffbias += numtrailingsyms(aligned[0],'_') + numtrailingsyms(aligned[1],'_')
for l in lines: # Read in lines and extract transformation rules from pairs
#fields out of order in 2023
#lemma, form, msd = l.split(u'\t')
lemma, msd, form = l.split(u'\t')
#reverse strings if there's a prefix bias
if prefbias > suffbias:
lemma = lemma[::-1]
form = form[::-1]
prules, srules = prefix_suffix_rules_get(lemma, form)
#keep track of all possible rules for each msd
if msd not in allprules and len(prules) > 0:
allprules[msd] = {}
if msd not in allsrules and len(srules) > 0:
allsrules[msd] = {}
#keep a count for how often each rule is used
for r in prules:
if (r[0],r[1]) in allprules[msd]:
allprules[msd][(r[0],r[1])] = allprules[msd][(r[0],r[1])] + 1
else:
allprules[msd][(r[0],r[1])] = 1
for r in srules:
if (r[0],r[1]) in allsrules[msd]:
allsrules[msd][(r[0],r[1])] = allsrules[msd][(r[0],r[1])] + 1
else:
allsrules[msd][(r[0],r[1])] = 1
#print(f'msds: {len(trainmsds)}')
# run eval on dev
#evallines = [line.strip() for line in open(path + lang.split("_")[0] + ".dev", "r") if line != '\n']
evallines = [line.strip() for line in open(path + lang + ".dev", "r") if line != '\n']
if TEST:
evallines = [line.strip() for line in open(path + lang.split("_")[0] + ".gold", "r") if line != '\n']
num_seenlemma_correct,num_seenmsd_correct,num_seenneither_correct = 0,0,0
num_seenlemma_guesses,num_seenmsd_guesses,num_seenneither_guesses = 0,0,0
numcorrect,numguesses = 0,0
if OUTPUT:
if not TEST:
outfile = open(lang + ".dev", "w")
else:
outfile = open(lang + ".test", "w")
for l in evallines:
#lemma, correct, msd, = l.split(u'\t')
lemma, msd, correct, = l.split(u'\t')
#correct,msd,lemma = l.split(u'\t')
# lemma, msd, = l.split(u'\t')
if prefbias > suffbias:
lemma = lemma[::-1]
outform = apply_best_rule(lemma, msd, allprules, allsrules)
if prefbias > suffbias:
outform = outform[::-1]
lemma = lemma[::-1]
if lemma in trainlemmas:
num_seenlemma_guesses += 1
if outform == correct:
num_seenlemma_correct += 1
elif msd in trainmsds:
num_seenmsd_guesses += 1
if outform == correct:
num_seenmsd_correct += 1
else:
num_seenneither_guesses += 1
if outform == correct:
num_seenneither_correct += 1
if OUTPUT:
outfile.write(lemma + "\t" + outform + "\t" + msd + "\n")
if OUTPUT:
outfile.close()
numlang += 1
if num_seenlemma_guesses:
percentcorrect_seenlemma = num_seenlemma_correct/num_seenlemma_guesses
totalavg_seenlemma += percentcorrect_seenlemma
numlang_seenlemma += 1
else:
percentcorrect_seenlemma = 0
if num_seenmsd_guesses:
percentcorrect_seenmsd = num_seenmsd_correct/num_seenmsd_guesses
totalavg_seenmsd += percentcorrect_seenmsd
numlang_seenmsd += 1
else:
percentcorrect_seenmsd = 0
if num_seenneither_guesses:
percentcorrect_seenneither = num_seenneither_correct/num_seenneither_guesses
totalavg_seenneither += percentcorrect_seenneither
numlang_seenneither += 1
else:
percentcorrect_seenneither = 0
numcorrect = num_seenlemma_correct + num_seenmsd_correct + num_seenneither_correct
numguesses = num_seenlemma_guesses + num_seenmsd_guesses + num_seenneither_guesses
totalavg += numcorrect/numguesses
print("%s:\t%s\t%s\t%s\t%s\t\t%s\t%s\t%s\t%s" % (lang, round(100*numcorrect/numguesses,3), round(100*percentcorrect_seenlemma,3), round(100*percentcorrect_seenmsd,3), round(100*percentcorrect_seenneither,3), numguesses, num_seenlemma_guesses, num_seenmsd_guesses, num_seenneither_guesses))
print("Average accuracy total\t\t", round(100*totalavg/numlang,3))
if numlang_seenlemma:
print("Average accuracy seen lemmas\t", round(100*totalavg_seenlemma/numlang_seenlemma,3))
if numlang_seenmsd:
print("Average accuracy seen msds\t", round(100*totalavg_seenmsd/numlang_seenmsd,3))
if numlang_seenneither:
print("Average accuracy seen neither\t", round(100*totalavg_seenneither/numlang_seenneither,3))
if __name__ == "__main__":
main(sys.argv)