def update(weights, predseq, labelseq, postagseq, vecs1, vecs2, sent):
for i in range(len(predseq)):
true = labelseq[i]
pred = predseq[i]
pos = postagseq[i]
vec1 = vecs1[i]
vec2 = vecs2[i]
if i == 0:
prev_true = '*'
prev_pred = '*'
else:
prev_true = labelseq[i - 1]
prev_pred = predseq[i - 1]
if true != pred:
true_feats = features.extract(sent[i], true, prev_true, pos, vec1,
vec2)
for feat in true_feats:
if feat in weights:
weights[feat] += step
pred_feats = features.extract(sent[i], pred, prev_pred, pos, vec1,
vec2)
for feat in pred_feats:
if feat in weights:
weights[feat] -= step
return weights
def ikProblemToFeatures(ikproblem,featureList):
"""Standard feature extractor for IKProblems"""
if isinstance(ikproblem,dict):
#assume it's a JSON object already
return features.extract(ikproblem,featureList)
elif isinstance(ikproblem,IKProblem):
jsonObj = ikproblem.toJson()
return features.extract(jsonObj,featureList)
elif isinstance(ikproblem,IKObjective):
return ikObjectiveToFeatures(ikproblem,featureList)
else:
assert hasattr(ikproblem,'__iter__'),"IK problem must either be an IKProblem, single IKObjective, or a list"
return sum([ikObjectiveToFeatures(o,f) for o,f in zip(ikproblem,featureList)],[])
def testPOSBIGRAMS(self):
text = "I love you."
results = features.extract(text, [Features.POSBIGRAMS], False)
result = results[Features.POSBIGRAMS]
self.assertEqual(1, result['(VBP)-(PRP)'])
self.assertEqual(1, result['(PRP)-(VBP)'])
self.assertEqual(1, result['(PRP)-(.)'])
results = features.extract(text, [Features.POSBIGRAMS], True)
result = results[Features.POSBIGRAMS]
self.assertEqual(1/3.0, result['(VBP)-(PRP)'])
self.assertEqual(1/3.0, result['(PRP)-(VBP)'])
self.assertEqual(1/3.0, result['(PRP)-(.)'])
self.assertEqual(1, sum(result.values()))
def testPOSUnigrams(self):
text = "I love you."
results = features.extract(text, [Features.POSUNIGRAM], False)
result = results[Features.POSUNIGRAM]
self.assertEqual(1, result['VBP'])
self.assertEqual(2, result['PRP'])
self.assertEqual(1, result['.'])
results = features.extract(text, [Features.POSUNIGRAM], True)
result = results[Features.POSUNIGRAM]
self.assertEqual(1/4.0, result['VBP'])
self.assertEqual(2/4.0, result['PRP'])
self.assertEqual(1/4.0, result['.'])
self.assertEqual(1, sum(result.values()))
def update(weights, predseq, labelseq, sent, postagseq, info, ad):
for i in range(len(predseq) + 1):
if i == len(predseq):
word = ''
pos = ''
true = '<STOP>'
pred = '<STOP>'
else:
word = sent[i]
true = labelseq[i]
pred = predseq[i]
pos = postagseq[i]
if i == 0:
prev_true = '*'
prev_pred = '*'
else:
prev_true = labelseq[i-1]
prev_pred = predseq[i-1]
if true != pred or i == len(predseq) and prev_true != prev_pred:
true_feats = extract(word, true, prev_true, pos, info)
pred_feats = extract(word, pred, prev_pred, pos, info)
up = set(true_feats).difference(set(pred_feats))
down = set(pred_feats).difference(set(true_feats))
# ADAGRAD
for u in up:
if u in ad:
ad[u] += 1
for d in down:
if d in ad:
ad[d] += 1
for u in up:
if u in weights:
if ad[u] > 0.0:
weights[u] += step/math.sqrt(ad[u]) # ADAGRAD
else:
weights[u] += step # ??
for d in down:
if d in weights:
if ad[d] > 0.0:
weights[d] -= step/math.sqrt(ad[d]) # ADAGRAD
else:
weights[d] -- step # ??
return weights
def test_level2(sourceword,target):
bestoutfn = "../L2output/{0}.{1}.best".format(sourceword, target)
oofoutfn = "../L2output/{0}.{1}.oof".format(sourceword, target)
bestoutfile = open(bestoutfn,'w')
oofoutfile = open(oofoutfn,'w')
level2_classifier = util_run_experiment.get_pickled_classifier(sourceword,target,'level2')
frd1,frd2,frd3,frd4 = sorted(list(get_four_friends(target))) ##Need 4 more features from level1.
classfrd1,classfrd2,classfrd3,classfrd4 = get_level1_classifiers(frd1,frd2,frd3,frd4,sourceword)
# finaldir = "../trialdata/alltrials/"
finaldir = "../finaltest"
problems = util_run_experiment.get_test_instances(finaldir, sourceword)
for problem in problems:
level1_features = features.extract(problem)
answer_frd1 = classfrd1.classify(level1_features)
answer_frd2 = classfrd2.classify(level1_features)
answer_frd3 = classfrd3.classify(level1_features)
answer_frd4 = classfrd4.classify(level1_features)
level2_features = train_extracted_level2.extend_features(level1_features,(answer_frd1,answer_frd2,answer_frd3,answer_frd4),frd1,frd2,frd3,frd4)
level2_answer = level2_classifier.classify(level2_features)
level2_dist = level2_classifier.prob_classify(level2_features)
oof_answers = util_run_experiment.topfive(level2_dist)
print(output_one_best(problem, target, level2_answer), file=bestoutfile)
print(output_five_best(problem, target, oof_answers),
file=oofoutfile)
def main():
parser = argparse.ArgumentParser(description='clwsd')
parser.add_argument('--sourceword', type=str, nargs=1, required=True)
parser.add_argument('--targetlang', type=str, nargs=1, required=True)
parser.add_argument('--classifier', type=str, nargs=1, required=False)
args = parser.parse_args()
all_target_languages = "de es fr it nl".split()
assert args.targetlang[0] in all_target_languages
target = args.targetlang[0]
sourceword = args.sourceword[0]
nltk.classify.megam.config_megam(bin='/usr/local/bin/megam')
classifier = get_maxent_classifier(sourceword, target)
fn = "../trialdata/alltrials/{0}.data".format(sourceword)
## XXX(alexr): fix later.
stanford.taggerhome = "/home/alex/software/stanford-postagger-2012-11-11"
problems = extract_wsd_problems(fn)
gold_answers = read_gold.get_gold_answers(sourceword, target)
for problem in problems:
featureset = features.extract(problem)
answer = classifier.classify(featureset)
print(problem.tokenized)
print(answer)
label = gold_answers[problem.instance_id]
print("CORRECT" if label == answer else "WRONG", end=" ")
print("should be:", label)
def main():
parser = argparse.ArgumentParser(description='clwsd')
parser.add_argument('--sourceword', type=str, required=True)
parser.add_argument('--targetlang', type=str, required=True)
parser.add_argument('--taggerhome', type=str, required=True)
args = parser.parse_args()
all_target_languages = "de es fr it nl".split()
assert args.targetlang in all_target_languages
target = args.targetlang
sourceword = args.sourceword
stanford.taggerhome = args.taggerhome
gold_answers = read_gold.get_gold_answers(sourceword, target)
instances = get_training_data(sourceword, target)
print("... training ...")
nltk.classify.megam.config_megam(bin='/usr/local/bin/megam')
classifier = MaxentClassifier.train(instances, trace=0, algorithm='megam')
print("LABELS", classifier.labels())
## with open("../eval/{0}.output".format(sourceword), "w") as outfile:
fn = "../trialdata/alltrials/{0}.data".format(sourceword)
problems = extract_wsd_problems(fn)
for problem in problems:
featureset = features.extract(problem)
answer = classifier.classify(featureset)
print(output_one_best(problem, target, answer))
label = gold_answers[problem.instance_id]
print("CORRECT" if label == answer else "WRONG")
print("distribution was...")
dist = classifier.prob_classify(featureset)
for key in dist.samples():
print(" ", key, dist.prob(key))
def main():
with open('model.pickle') as f:
vectorizer, model = pickle.load(f)
for hyp1, hyp2, ref in sentences():
f1 = features.extract(ref, hyp1)
f2 = features.extract(ref, hyp2)
f = features.diff(f2, f1)
if f['min_match'] == f['min_match2'] == f['min_match3'] == 0:
print 0
continue
score = model.predict(vectorizer.transform((f,))) # w . (f_2 - f_1)
if score > 0:
print 1
else:
print -1
def mrf_optimize(problem):
"""Build the MRF and do the optimization!!"""
featureset = features.extract(problem)
for lang in all_target_languages:
classifier = classifiers[lang]
unary = unary_penalty_table(classifier, featureset)
print(unary)
langnode = Node(lang, unary)
## get all the combinations of nodes.
langpairs = list(itertools.combinations(all_target_languages, 2))
## create an edge for each language pair.
for l1, l2 in langpairs:
print("building edges for", l1, l2)
edgepotentials = {}
for val1 in possible_values(l1):
for val2 in possible_values(l2):
cooccurrence = cooccurrences[(l1,l2)]
joint = cooccurrence.lookup_joint(l1, val1, l2, val2)
# negative logprob of the joint probability. Definitely the best
# edge potential, for sure.
edgepotentials[(val1,val2)] = -math.log(joint, 2)
Edge(l1, l2, edgepotentials)
## XXX how many iterations?
print("mrf solving!!!")
answers, oof_answers = beliefprop(10)
print("mrf solved!!!")
return answers, oof_answers
def get_training_data_from_extracted(sourceword, targetlang):
"""Return a list of (featureset, label) for training."""
out = []
problems = []
fn = "../trainingdata/{0}.{1}.train".format(sourceword, targetlang)
with open(fn) as infile:
lines = infile.readlines()
lines = [line.strip() for line in lines]
contexts = [line for line in lines[0::3]]
indices = [int(line) for line in lines[1::3]]
labelss = [line.split(",") for line in lines[2::3]]
assert len(contexts) == len(labelss) == len(indices)
answers = []
for context, index, labels in zip(contexts, indices, labelss):
problem = WSDProblem(sourceword, context,
testset=False, head_index=index)
for label in labels:
if label == '': continue
problems.append(problem)
answers.append(label)
for problem,answer in zip(problems, answers):
featureset = features.extract(problem)
label = answer
assert(type(label) is str)
out.append((featureset, label))
#print("###the features are: \n{}".format(featureset))
#input()
return out
def extract_all_features(formatted_corpus):
sent_cnt = 0
y_symbols = [] # Our array of transistions
X_dict = list() # Our matrix
for sentence in formatted_corpus:
sent_cnt += 1
if sent_cnt % 1000 == 0:
a = 1
# print(sent_cnt, 'sentences on', len(formatted_corpus), flush=True)
stack = []
queue = list(sentence)
graph = {}
graph['heads'] = {}
graph['heads']['0'] = '0'
graph['deprels'] = {}
graph['deprels']['0'] = 'ROOT'
while queue:
x = features.extract(stack, queue, graph, FEATURE_NAMES, sentence)
X_dict.append(x)
stack, queue, graph, trans = reference(stack, queue, graph)
y_symbols.append(trans)
stack, graph = transition.empty_stack(stack, graph)
# Poorman's projectivization to have well-formed graphs.
for word in sentence:
word['head'] = graph['heads'][word['id']]
# print(y_symbols)
# print(graph)
return X_dict, y_symbols
def read_dataset(filename):
X, y = [], []
points = features.extract(filename)
for f, v in points:
X.append(f.todict())
y.append(v)
return X, y
def getTrainDS():
root = './data'
X = np.mat([0]*(len(markerList)))
y = np.array([])
for lang in langs:
for scDir in os.listdir(os.path.join(root, lang)):
sc = getSC(os.path.join(root, lang, scDir))
feature = extract(sc)
feature = np.asmatrix(feature)
X = np.append(X, feature, axis=0)
X = np.asmatrix(X)
y = np.append(y, [lang])
print "[SUCCESS] Extracted", scDir
del feature
X = X[1:]
print "Shuffling datasets."
X, y = shuffle(X, y)
print "[SUCCES] Shuffled datasets"
print "Splitting and saving datasets."
ds = splitDS(X, y)
print "[SUCCESS] Splitted datasets."
saveDS(ds)
print "[SUCCESS] Saved datasets."
del X, y
def testLetterUnigram(self):
text = "ekin"
results = features.extract(text,[Features.LETTERUNIGRAM],False)
result = results[Features.LETTERUNIGRAM]
self.assertEqual(1, result[ord('e')-ord('a')])
self.assertEqual(1, result[ord('k')-ord('a')])
self.assertEqual(1, result[ord('i')-ord('a')])
self.assertEqual(1, result[ord('n')-ord('a')])
self.assertEqual(4, sum(result))
results = features.extract(text,[Features.LETTERUNIGRAM],True)
result = results[Features.LETTERUNIGRAM]
self.assertEqual(1/4.0, result[ord('e')-ord('a')])
self.assertEqual(1/4.0, result[ord('k')-ord('a')])
self.assertEqual(1/4.0, result[ord('i')-ord('a')])
self.assertEqual(1/4.0, result[ord('n')-ord('a')])
self.assertEqual(4/4.0, sum(result))
text = ""
results = features.extract(text,[Features.LETTERUNIGRAM],False)
result = results[Features.LETTERUNIGRAM]
self.assertEqual(0, sum(result))
results = features.extract(text,[Features.LETTERUNIGRAM],True)
result = results[Features.LETTERUNIGRAM]
self.assertEqual(0, sum(result))
text = "eee"
results = features.extract(text,[Features.LETTERUNIGRAM],False)
result = results[Features.LETTERUNIGRAM]
self.assertEqual(3, result[ord('e')-ord('a')])
self.assertEqual(3, sum(result))
results = features.extract(text,[Features.LETTERUNIGRAM],True)
result = results[Features.LETTERUNIGRAM]
self.assertEqual(1.0, result[ord('e')-ord('a')])
self.assertEqual(1.0, sum(result))
text = "ee k ee"
results = features.extract(text,[Features.LETTERUNIGRAM],False)
result = results[Features.LETTERUNIGRAM]
self.assertEqual(4, result[ord('e')-ord('a')])
self.assertEqual(1, result[ord('k')-ord('a')])
self.assertEqual(5, sum(result))
results = features.extract(text,[Features.LETTERUNIGRAM],True)
result = results[Features.LETTERUNIGRAM]
self.assertEqual(4/5.0, result[ord('e')-ord('a')])
self.assertEqual(1/5.0, result[ord('k')-ord('a')])
self.assertEqual(5/5.0, sum(result))
text = "\nn"
results = features.extract(text, [Features.LETTERUNIGRAM], False)
result = results[Features.LETTERUNIGRAM]
self.assertEqual(1.0, result[ord('n')-ord('a')])
self.assertEqual(1.0, sum(result))
def main(args: argparse.Namespace) -> None:
with open(args.data, "r") as src:
with open(args.features, "w") as out_file:
for line in src:
line = word_tokenize(line.replace(":", "_"))
feature_list = features.extract(line)
for feature in feature_list:
print("\t".join(feature), file=out_file)
print("", file=out_file)
def get_score(word, current_tag, prev_tag, postag, weights, info):
score = 0.0
features_list = extract(word, current_tag, prev_tag, postag, info)
for feature in features_list:
if feature in weights:
score += weights[feature]
return score
def get_score(word, current_tag, prev_tag, pos, v1, v2, weights):
score = 0.0
features_list = extract(word, current_tag, prev_tag, pos, v1, v2)
for feature in features_list:
if feature in weights:
score += weights[feature]
return score, features_list
def get_score(word, current_tag, prev_tag, pos, v1, v2, weights):
score = 0.0
features_list = extract(word, current_tag, prev_tag, pos, v1, v2)
for feature in features_list:
if feature in weights:
score += weights[feature]
return score, features_list
def playing(colrow, g): # takes a board position and adds a players move
col, row = colrow[0], colrow[1:3]
row = BOARD - int(row)
col = alphabet.index(col)
tmp = g[0].copy()
g[0] = g[1]
g[1] = tmp
g = features.extract(g, row, col, BOT)
return prediction(g)
def ikProblemToFeatures(ikproblem, featureList):
"""Standard feature extractor for IKProblems"""
if isinstance(ikproblem, dict):
#assume it's a JSON object already
return features.extract(ikproblem, featureList)
elif isinstance(ikproblem, IKProblem):
jsonObj = ikproblem.toJson()
return features.extract(jsonObj, featureList)
elif isinstance(ikproblem, IKObjective):
return ikObjectiveToFeatures(ikproblem, featureList)
else:
assert hasattr(
ikproblem, '__iter__'
), "IK problem must either be an IKProblem, single IKObjective, or a list"
return sum([
ikObjectiveToFeatures(o, f)
for o, f in zip(ikproblem, featureList)
], [])
def read_dataset(filename, use_text_features):
if use_text_features:
mask = set(['meta', 'text'])
else:
mask = set(['meta'])
X, y = [], []
points = list(features.filter(features.extract(filename), mask))
for f, v in points:
X.append(f.todict())
y.append(v)
return X, y
def get_training_data_from_extracted(sourceword, targetlang):
"""Return a list of (featureset, label) for training."""
frd1, frd2, frd3, frd4 = sorted(list(get_four_friends(targetlang))) ##Get other four languages.
##Get the intersection of four training sentences.
tool_class = Occurrence(sourceword, frd1, frd2)
intersection = tool_class.get_common_four_sents(sourceword, frd1, frd2, frd3, frd4)
out = []
problems = []
fn = "../trainingdata/{0}.{1}.train".format(sourceword, targetlang)
with open(fn) as infile:
lines = infile.readlines()
lines = [line.strip() for line in lines]
contexts = [line for line in lines[0::3]]
indices = [int(line) for line in lines[1::3]]
labelss = [line.split(",") for line in lines[2::3]]
assert len(contexts) == len(labelss) == len(indices)
print("the length of them...", len(contexts), len(indices), len(labelss))
# input()
answers = []
extention = []
for context, index, labels in zip(contexts, indices, labelss):
sentence_id = context + "####" + str(index)
if (
sentence_id in intersection
): ##If this sentence also appears in 4 other languages, we can use more features...
problem = WSDProblem(sourceword, context, testset=False, head_index=index)
more_featuress = intersection[sentence_id]
# print(more_featuress)
for more_feature in more_featuress:
for label in labels:
if label == "":
continue
problems.append(problem)
# more_features = intersection[context]
extention.append(more_feature)
answers.append(label)
print("###intersection for five languages....{}\n".format(len(extention)))
for problem, answer, more_feature in zip(problems, answers, extention):
featureset = features.extract(problem)
featureset = extend_features(featureset, more_feature, frd1, frd2, frd3, frd4)
label = answer
assert type(label) is str
# print("=================@@@@features {}\n@@@@label{}\n".format(featureset,label))
out.append((featureset, label))
print("###Length of the output should be the same{}\n".format(len(out)))
return out
def extract_all_train_features(sents, tagseqs, postagseqs, info):
featset = set([])
i = 0
for sent in sents:
sys.stderr.write(str(i) + "\r")
j = 0
for word in sent:
tag = tagseqs[i][j]
postag = postagseqs[i][j]
if j == 0: # first position
prev = '*'
else:
prev = tagseqs[i][j-1]
featset.update(extract(word, tag, prev, postag, info)) # get a list of all features possible
j += 1
# features for the last label
featset.update(extract('', '<STOP>', tag, '', info))
i += 1
featlist = list(featset)
for f in featlist:
print f
return featlist
def extract_features(self):
"""
This method is used to extract features (A,B,D).
"""
returnVars = features.extract(self.original_image, self.contour_mask,
self.contour)
if len(returnVars) == 0:
self.feature_set = returnVars
else:
self.feature_set = returnVars[0]
self.asymmetry_horizontal = returnVars[1]
self.asymmetry_vertical = returnVars[2]
self.warp_img_segmented = returnVars[3]
def get_training_data(sourceword, target):
"""Return a list of (featureset, label) for training."""
out = []
## map from id to labels
gold_answers = read_gold.get_gold_answers(sourceword, target)
problems = get_training_problems(sourceword)
## now collate them.
for problem in problems:
theid = problem.instance_id
featureset = features.extract(problem)
label = gold_answers[theid]
out.append((featureset, label))
return out
def convert_to_features(data_part) -> np.ndarray:
"""This converts a given group to features"""
current_features = Features()
feature_list = list()
last_features = current_features
for row in data_part.itertuples():
row_data = Row(row)
current_features.update(row_data)
feature_list.append(
features.extract(row_data, current_features, last_features))
last_features = current_features.snapshot()
return np.array(feature_list)
def get_training_data_from_extracted(sourceword, targetlang):
"""Return a list of (featureset, label) for training."""
frd1,frd2,frd3,frd4 = sorted(list(get_four_friends(targetlang))) ##Get other four languages.
classfrd1,classfrd2,classfrd3,classfrd4 = get_level1_classifiers(frd1,frd2,frd3,frd4,sourceword)
##Get the intersection of four training sentences.
tool_class = Occurrence(sourceword,frd1,frd2)
out = []
problems = []
fn = "../trainingdata/{0}.{1}.train".format(sourceword, targetlang)
with open(fn) as infile:
lines = infile.readlines()
lines = [line.strip() for line in lines]
contexts = [line for line in lines[0::3]]
indices = [int(line) for line in lines[1::3]]
labelss = [line.split(",") for line in lines[2::3]]
assert len(contexts) == len(labelss) == len(indices)
print("the length of them...",len(contexts),len(indices),len(labelss))
#input()
answers = []
for context, index, labels in zip(contexts, indices, labelss):
problem = WSDProblem(sourceword, context,
testset=False, head_index=index)
#print(more_featuress)
for label in labels:
if label == '': continue
problems.append(problem)
#more_features = intersection[context]
answers.append(label)
for problem,answer in zip(problems, answers):
level1_features = features.extract(problem)
answer_frd1 = classfrd1.classify(level1_features)
answer_frd2 = classfrd2.classify(level1_features)
answer_frd3 = classfrd3.classify(level1_features)
answer_frd4 = classfrd4.classify(level1_features)
level2_features = extend_features(level1_features,(answer_frd1,answer_frd2,answer_frd3,answer_frd4),frd1,frd2,frd3,frd4)
label = answer
assert(type(label) is str)
#print("=================@@@@features {}\n@@@@label{}\n".format(featureset,label))
out.append((level2_features, label))
print("###Length of the output should be the same{}\n".format(len(out)))
return out
def main():
print(os.getcwd())
print("Create Dataset")
signal, y = dataset()
df_y = pd.DataFrame(data=y, columns=['genre'])
# construct features
print("Feature Extraction")
df_x = pd.DataFrame()
for i in range(0, len(signal)):
print("number " + str(i))
new_x = pd.DataFrame(features.extract(signal[i]), index=[i])
df_x = df_x.append(new_x)
saveFeature(df_x, df_y)
def disambiguate_words(words):
"""Given a list of words/lemmas, return a list of disambiguation answers for
them."""
classifiers = [classifier_for(word, nonnull=True) for word in words]
answers = []
for i in range(len(words)):
faketagged = [(w,None) for w in words]
feat = features.extract(faketagged, i)
classif = classifiers[i]
ans = classif.classify(feat)
if ans == UNTRANSLATED:
ans = mfs_translation(words[i])
print("MFS!!!", words[i], "==>", ans)
answers.append(ans)
return [str(ans) for ans in answers]
def main():
setup_logging()
for fn in sys.argv[1:] or ["input.png"]:
im = cv.LoadImage(fn)
fts = extract(im)
pfn = fn + "-features.dat"
info("Storing feature pickle in %s", pfn)
dump(fts, file(pfn, "wb"))
for l, layer in enumerate(fts):
for fname, fval in layer.items():
ffn = "%s-feat-%d-%s.png" % (fn, l, fname)
info("Rendering feature %s", ffn)
mat2pil(fval).save(ffn)
def update(weights, predseq, labelseq, postagseq, vecs1, vecs2, sent):
for i in range(len(predseq)):
true = labelseq[i]
pred = predseq[i]
pos = postagseq[i]
vec1 = vecs1[i]
vec2 = vecs2[i]
if i == 0:
prev_true = '*'
prev_pred = '*'
else:
prev_true = labelseq[i-1]
prev_pred = predseq[i-1]
if true != pred:
true_feats = features.extract(sent[i], true, prev_true, pos, vec1, vec2)
for feat in true_feats:
if feat in weights:
weights[feat] += step
pred_feats = features.extract(sent[i], pred, prev_pred, pos, vec1, vec2)
for feat in pred_feats:
if feat in weights:
weights[feat] -= step
return weights
def prediction(gamemove, col):
predictmove = gamemove[None, :]
resultmatrix = model.predict(predictmove)
resultmatrix = np.reshape(resultmatrix, [BOARD, BOARD])
result = np.unravel_index(resultmatrix.argmax(), resultmatrix.shape)
while gamemove[4][result[0]][result[1]] == 0: # if move is illegal
resultmatrix[result[0]][result[1]] = 0 # delete from softmax
result = np.unravel_index(resultmatrix.argmax(),
resultmatrix.shape) #find second best option
resultmatrix = np.round_(resultmatrix, 1) * 10
print(alphabet[result[1] - BOARD], BOARD - result[0])
tmp = gamemove[0].copy()
gamemove[0] = gamemove[1]
gamemove[1] = tmp
gamemove = features.extract(gamemove, result[0], result[1], col)
return gamemove, resultmatrix
def extract_features(formatted_corpus, feature_names, training=True, model=None):
non_proj = []
X_1 = []
y_1 = []
sent_cnt = 0
for sentence in formatted_corpus:
sent_cnt += 1
if sent_cnt % 1000 == 0:
print(sent_cnt, 'sentences on', len(formatted_corpus), flush=True)
stack = []
queue = list(sentence)
graph = {}
graph['heads'] = {}
graph['heads']['0'] = '0'
graph['deprels'] = {}
graph['deprels']['0'] = 'ROOT'
transitions = []
feats = []
while queue:
feats.append(features.extract(stack, queue, graph, feature_names, sentence))
stack, queue, graph, trans = reference(stack, queue, graph)
transitions.append(trans)
stack, graph = transition.empty_stack(stack, graph)
X_1.extend(feats)
y_1.extend(transitions)
#print('Equal graphs:', transition.equal_graphs(sentence, graph))
if not transition.equal_graphs(sentence, graph):
non_proj.append(sentence)
# Poorman's projectivization to have well-formed graphs.
for word in sentence:
word['head'] = graph['heads'][word['id']]
#print(transitions)
#print(graph)
#print(len(non_proj))
#s = sorted(non_proj, key=lambda x: len(x))
#print([x['form'] for x in s[0]])
#for x in non_proj:
# print(len(x))
# print(x)
return (X_1, y_1)
def prob_disambiguate_words(words):
"""Given a list of words/lemmas, return a list of disambiguation answers for
them -- return a list of lists, where each sublist is ordered in decreasing
probability."""
classifiers = [classifier_for(word, nonnull=True) for word in words]
answers = []
for i in range(len(words)):
faketagged = [(w,None) for w in words]
feat = features.extract(faketagged, i)
classif = classifiers[i]
## get all possible options, sorted in wrong order
dist = classif.prob_classify(feat)
options = [(dist.prob(samp), samp) for samp in dist.samples()]
options.sort(reverse=True)
myanswers = [str(lex) for (prob, lex) in options
if prob > 0.01 ]
print(myanswers)
answers.append(myanswers)
return answers
def test(file_name):
# file_name="combined_nomeal.csv"
with open("Guassian_model.pkl", 'rb') as file:
Guassian_model = pickle.load(file)
with open("pca_model.pkl", 'rb') as file:
pca = pickle.load(file)
test_data = pd.read_csv(file_name, header=None)
print("---")
fm = extract(test_data)
print("---")
sc = StandardScaler()
test_data_set = sc.fit_transform(fm)
pca_dataset = pca.fit_transform(test_data_set)
gaussianNB_pred = Guassian_model.predict(pca_dataset)
print("Classes of your given test " + str(gaussianNB_pred))
np.savetxt("output.csv", gaussianNB_pred, delimiter=",", fmt='%d')
print("predicted class labels are saved in output.csv file")
def perform_prediction(split_test_sentences, feature_names, classifier, model):
print("Extracting features from test Corpus")
X_matrix_test = [] # The matrix of feature vectors
Y_vec_test = [] # The vector of predicted outputs
# We iterate over all sentences that are in the corpus
for sentence in split_test_sentences:
stack = [] # Stack of tokens that will be manipulated with shift, reduce, left-arc, and right-arc
queue = list(sentence) # Queue of input tokens that will be cycles through
graph = {} # The dependency graph that we will go store our dependency arcs in
graph['heads'] = {} # Create a dictionary inside graph for heads of each sentence
graph['heads']['0'] = '0' # The first element in the heads dictionary inside the graph dictionary is '0'
graph['deprels'] = {} # Define another dictionary with dependency relations stored inside, again inside graph
graph['deprels']['0'] = 'ROOT' # Make the first element in the deprel dictionary the 'ROOT' keyword
while queue:
x_elem = features.extract(stack, queue, graph, feature_names, sentence)
x_predictable = features.vec.transform(x_elem)
# X_matrix_test.extend(x_elem)
# stack, queue, graph, y_elem = dparser.reference(stack, queue, graph)
# WE DO NOT USE Y_VEC_TEST
# Y_vec_test.append(y_elem)
predicted_transition = classifier.predict(x_predictable)
stack, queue, graph, transition = parse_ml(stack, queue, graph, predicted_transition)
def extract_features(csv_path, csv_file):
data = []
f_temp = []
with open(csv_path, "r") as data_file: #open the files
for row in data_file:
strsplit = row.split(',')
strsplit = list(map(float, strsplit))
data.append(strsplit)
no_of_data_per_window = int(sampling_rate * window_size)
no_of_windows = len(data) // no_of_data_per_window * 2 - 1
#count = 0
for i in range(no_of_windows):
window_slice_data = []
for j in range(no_of_data_per_window):
window_slice_data.append(data[i * no_of_data_per_window // 2 + j])
#count += 1
#print(count)
window_slice_data = np.asarray(window_slice_data)[:, 0:3].tolist()
f_temp.append(features.extract(window_slice_data))
l_temp = np.full(len(f_temp), table[csv_file])
return f_temp, l_temp
def getData():
if (os.path.exists('df_no_index_10mfcc.csv')):
df = pd.read_csv('df_no_index_10mfcc.csv')
df_y = df['genre']
df_x = df.drop('genre', axis=1)
return df_x, df_y
# features haven't been created
print("Create Dataset")
signal, y = saveDataset.dataset()
df_y = pd.Series(data=y)
# construct features
print("Feature Extraction")
df_x = pd.DataFrame()
for i in range(len(signal)):
if (i % 50 == 0):
print('audio {}'.format(i + 1))
new_x = pd.DataFrame(features.extract(signal[i]), index=[i])
df_x = df_x.append(new_x)
saveDataset.saveFeature(df_x, pd.DataFrame(df_y, columns=['genre']))
return df_x, df_y
def main():
parser = util_run_experiment.get_argparser()
args = parser.parse_args()
assert args.targetlang in all_target_languages
assert args.sourceword in all_words
targetlang = args.targetlang
sourceword = args.sourceword
trialdir = args.trialdir
stanford.taggerhome = args.taggerhome
print("Loading and tagging test problems...")
problems = util_run_experiment.get_test_instances(trialdir, sourceword)
print("OK loaded and tagged.")
## classifier = get_maxent_classifier(sourceword, targetlang)
classifier = get_pickled_classifier(sourceword, targetlang, "level1")
if not classifier:
print("Couldn't load pickled L1 classifier?")
return
print("Loaded pickled L1 classifier!")
bestoutfn = "../L1output/{0}.{1}.best".format(sourceword, targetlang)
oofoutfn = "../L1output/{0}.{1}.oof".format(sourceword, targetlang)
with open(bestoutfn, "w") as bestoutfile, \
open(oofoutfn, "w") as oofoutfile:
for problem in problems:
featureset = features.extract(problem)
answer = classifier.classify(featureset)
dist = classifier.prob_classify(featureset)
oof_answers = util_run_experiment.topfive(dist)
print(output_one_best(problem, targetlang, answer),
file=bestoutfile)
print(output_five_best(problem, targetlang, oof_answers),
file=oofoutfile)
def generate(track):
blob = extract(track.source, track.details_file)
features.extract(blob, track)
update_tags(blob, track)
# collect val and time sequence from addresses
dirPath = 'test_addr'
addrs = os.listdir(dirPath)
p = connectPSQL(psql)
'''
for addr in addrs:
if addr[0]!='d':
color.pImportant('addr file: '+addr)
full_path = os.path.join(dirPath,addr)
data_file = 'test_'+addr.split('.')[0].split('_')[1]+'_database.csv'
data_file = os.path.join('result',data_file)
in_csv = collectTxnIn(p,full_path)
out_csv = collectTxnOut(p,full_path)
deal_sql.deal_feature(in_csv, out_csv, data_file)
feature.extract(data_file)
os.rename(full_path,os.path.join(dirPath,'done-'+addr))
'''
addr = 'add_ponzi_train.csv'
color.pImportant('addr file: ' + addr)
full_path = os.path.join(dirPath, addr)
data_file = 'test_' + addr.split('.')[0].split('_')[1] + '_database.csv'
data_file = os.path.join('result', data_file)
in_csv = collectTxnIn(p, full_path)
out_csv = collectTxnOut(p, full_path)
deal_sql.deal_feature(in_csv, out_csv, data_file)
feature.extract(data_file)
p.close()
#################################
meal.to_csv("combined_meal.csv", index=False, header=False)
nomeal.to_csv("combined_nomeal.csv", index=False, header=False)
#
###########################################
X = [29 - i for i in range(0, 30)]
#for i in range(10) :
# plt.plot(X,meal.iloc[73,:],marker="X")
#
#print("AFTER FUNCTION")
###call function
feature_Matrix_Meal = pd.DataFrame()
feature_Matrix_Meal = extract(meal)
feature_Matrix_NoMeal = pd.DataFrame()
feature_Matrix_NoMeal = extract(nomeal)
#
#feature_Matrix_Meal['class']=1
#feature_Matrix_NoMeal['class']=0
#
from sklearn.decomposition import PCA
from sklearn.preprocessing import StandardScaler
sc = StandardScaler()
train_meal = sc.fit_transform(feature_Matrix_Meal)
train_nomeal = sc.fit_transform(feature_Matrix_NoMeal)
def ikObjectiveToFeatures(ikgoal,featureList):
"""Given an IKObjective instance and a list of features, returns a
feature descriptor as a list of floats"""
jsonObj = loader.toJson(ikgoal)
return features.extract(jsonObj,featureList)
import model
import performance
import dataset
if __name__ == "__main__":
# We begin by downloading the data. The data will be in the form of
# "events" data: each datapoint for each patient will be a recorded event.
X, Y = download.download()
# The event data is reformatted. This is done by selecting the given
# variables and transforming time-dependent events to a path.
X = reformat.reformat(X,
static_variables=["Age", "Gender"],
dynamic_variables=["Creatinine", "Glucose"])
# Now, we normalise the data.
X = normalise.normalise(X)
# We extract features from the input data.
features = features.extract(X)
# The dataset is now split into a training and testing set.
features_train, Y_train, features_test, Y_test = dataset.split(
features, Y, proportion=0.75)
# We now train the model with the selected features.
classifier = model.train(features_train, Y_train)
# We evaluate performance of the model now.
performance.evaluate(classifier, features_test, Y_test)
p = connectPSQL(psql)
times = [time.time()]
#for addr in addrs:
for i in range(1):
addr = 'new1.csv'
color.pImportant('addr file: ' + addr)
full_path = os.path.join(dirPath, addr)
in_csv = collectTxnIn(p, full_path)
out_csv = collectTxnOut(p, full_path)
times.append(time.time())
color.pImportant('collected all txns in ' + str(times[-1] - times[-2]))
data_file = addr.split('.')[0] + '_database.csv'
data_file = os.path.join('result', data_file)
deal_sql.deal_feature(in_csv, out_csv, data_file)
feature_file = feature.extract(data_file)
times.append(time.time())
color.pImportant('dealed all datas in ' + str(times[-1] - times[-2]))
color.pImportant('')
'''
feature_df = pd.read_csv(feature_file)
label_df = pd.read_csv(os.path.join(dirPath,label),header=None)
label_df.columns=['ponzi']
feature_df['ponzi'] = label_df['ponzi']
feature_df.to_csv(feature_file,index=None)
'''
p.close()
color.pImportant('total time used: ' + str(times[-1] - times[0]))
def ikObjectiveToFeatures(ikgoal, featureList):
"""Given an IKObjective instance and a list of features, returns a
feature descriptor as a list of floats"""
jsonObj = loader.toJson(ikgoal)
return features.extract(jsonObj, featureList)
for sentence in formatted_corpus:
sent_cnt += 1
if sent_cnt % 1000 == 0:
a = 1
# print(sent_cnt, 'sentences on', len(formatted_corpus), flush=True)
stack = []
queue = list(sentence)
graph = {}
graph['heads'] = {}
graph['heads']['0'] = '0'
graph['deprels'] = {}
graph['deprels']['0'] = 'ROOT'
while queue:
x = features.extract(stack, queue, graph, FEATURE_NAMES, sentence)
X_test = vec.transform(x)
predicted_trans_index = model.predict(X_test)[0]
predicted_trans = dict_classes[predicted_trans_index]
# Build new graph
stack, queue, graph, trans = execute_transition(
stack, queue, graph, predicted_trans)
# Save the predicted trans
y_predicted_symbols.append(trans)
stack, graph = transition.empty_stack(stack, graph)
def testLetterBigrams(self):
text = ""
results = features.extract(text,[Features.LETTERBIGRAMS],False)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(0, sum(result))
results = features.extract(text,[Features.LETTERBIGRAMS],True)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(0, sum(result))
text = "eee"
results = features.extract(text,[Features.LETTERBIGRAMS],False)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(2, result['ee'])
self.assertEqual(2, sum(result.values()))
results = features.extract(text,[Features.LETTERBIGRAMS],True)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(1.0, result['ee'])
self.assertEqual(1.0, sum(result.values()))
text = "eee eee e ee"
results = features.extract(text,[Features.LETTERBIGRAMS],False)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(5, result['ee'])
self.assertEqual(5, sum(result.values()))
results = features.extract(text,[Features.LETTERBIGRAMS],True)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(1.0, result['ee'])
self.assertEqual(1.0, sum(result.values()))
text = "eekke eee e eze zzkzk"
results = features.extract(text,[Features.LETTERBIGRAMS],False)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(3, result['ee'])
self.assertEqual(1, result['ek'])
self.assertEqual(1, result['kk'])
self.assertEqual(1, result['ke'])
self.assertEqual(1, result['ez'])
self.assertEqual(1, result['ze'])
self.assertEqual(2, result['zk'])
self.assertEqual(1, result['kz'])
self.assertEqual(1, result['zz'])
self.assertEqual(11+1, sum(result.values()))
results = features.extract(text,[Features.LETTERBIGRAMS],True)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(3/12.0, result['ee'])
self.assertEqual(1/12.0, result['ek'])
self.assertEqual(1/12.0, result['kk'])
self.assertEqual(1/12.0, result['ke'])
self.assertEqual(1/12.0, result['ez'])
self.assertEqual(1/12.0, result['ze'])
self.assertEqual(2/12.0, result['zk'])
self.assertEqual(1/12.0, result['kz'])
self.assertEqual(1/12.0, result['zz'])
self.assertEqual(1.0, sum(result.values()))
text = "ababababa"
results = features.extract(text,[Features.LETTERBIGRAMS],False)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(4, result['ab'])
self.assertEqual(4, result['ba'])
self.assertEqual(8, sum(result.values()))
results = features.extract(text,[Features.LETTERBIGRAMS],True)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(4/8.0, result['ab'])
self.assertEqual(4/8.0, result['ba'])
self.assertEqual(1.0, sum(result.values()))
text = "ab \nn \tt"
results = features.extract(text, [Features.LETTERBIGRAMS], False)
result = results[Features.LETTERBIGRAMS]
self.assertEqual(1.0, result['ab'])
self.assertEqual(1.0, sum(result.values()))
def testSaveAndLoad(self):
text = "ekin and ekin."
results = features.extract(text,
[Features.LETTERUNIGRAM, Features.LETTERBIGRAMS], False)
features.save(results, "test1")
returned = features.load("test1")
os.remove("test1")
self.assertEquals(results, returned)
self.assertEqual(2, returned[Features.LETTERUNIGRAM][ord('e')-ord('a')])
results = features.extract(text,
[Features.POSBIGRAMS, Features.LETTERBIGRAMS], True)
features.save(results, "test2")
returned = features.load("test2")
os.remove("test2")
self.assertEquals(results, returned)
results = features.extract(text,
[Features.LETTERUNIGRAM, Features.LETTERBIGRAMS, Features.POSUNIGRAM, Features.POSBIGRAMS], True)
features.save(results, "test3")
returned = features.load("test3")
os.remove("test3")
self.assertEquals(results, returned)
text = ""
results = features.extract(text,
[Features.LETTERUNIGRAM, Features.LETTERBIGRAMS, Features.POSUNIGRAM, Features.POSBIGRAMS], False)
features.save(results, "test4")
returned = features.load("test4")
os.remove("test4")
self.assertEquals(results, returned)
text = "\ttrying to test really long text\n\n \a and see if it still works or not \n 101 531 z3z3\n\t"
text += text;
text += text;
results = features.extract(text,
[Features.LETTERUNIGRAM, Features.LETTERBIGRAMS, Features.POSUNIGRAM, Features.POSBIGRAMS], False)
features.save(results, "test5")
returned = features.load("test5")
os.remove("test5")
self.assertEquals(results, returned)
text = ['test1', '222aaa', 'threee333', 't\test1'];
results['authors'] = [0, 0, 1, 1]
results['text'] = list();
for t in text:
results['text'].append(features.extract(t, [Features.LETTERUNIGRAM, Features.LETTERBIGRAMS, Features.POSUNIGRAM, Features.POSBIGRAMS], False))
features.save(results, "test6");
returned = features.load("test6");
os.remove("test6")
self.assertEqual(results['authors'], returned['authors'])
for idx in range(0,len(returned['text'])):
feat = features.extract(text[idx], [Features.LETTERUNIGRAM, Features.LETTERBIGRAMS, Features.POSUNIGRAM, Features.POSBIGRAMS], False)
self.assertEqual(feat, returned['text'][idx])
print(sent_cnt, 'sentences on', len(formatted_corpus), flush=True)
stack = []
queue = list(sentence)
graph = {}
graph['heads'] = {}
graph['heads']['0'] = '0'
graph['deprels'] = {}
graph['deprels']['0'] = 'ROOT'
transitions = []
#feature_names = ["stack_pos", "stack_word", "queue_pos", "queue_word", "can_ra", "can_la"]
while queue:
# print(queue)
x = features.extract(stack, queue, graph, feature_names, sentence)
X.append(
dict(zip(feature_names, x))
) #Nu producerar denna en lång lista, där varje element är en mening. Ändra till att vare ord är ett eget element
stack, queue, graph, trans = reference(stack, queue, graph)
transitions.append(trans)
y = trans
y_vector.append(y)
stack, graph = transition.empty_stack(stack, graph)
# print('Equal graphs:', transition.equal_graphs(sentence, graph))
# Poorman's projectivization to have well-formed graphs.
for word in sentence:
word['head'] = graph['heads'][word['id']]
X = []
y = []
for sentence in formatted_corpus:
sent_cnt += 1
#if sent_cnt % 1000 == 0:
#print(sent_cnt, 'sentences on', len(formatted_corpus), flush=True)
stack = []
queue = list(sentence)
graph = {}
graph['heads'] = {}
graph['heads']['0'] = '0'
graph['deprels'] = {}
graph['deprels']['0'] = 'ROOT'
transitions = []
while queue:
X.append(features.extract(stack, queue, graph, feature_names, sentence))
stack, queue, graph, trans = reference(stack, queue, graph)
transitions.append(trans)
y.append(trans)
stack, graph = transition.empty_stack(stack, graph)
#print('Equal graphs:', transition.equal_graphs(sentence, graph))
# Poorman's projectivization to have well-formed graphs.
for word in sentence:
word['head'] = graph['heads'][word['id']]
#print(transitions)
#print(graph)
vec = DictVectorizer(sparse=True)
modelFilename = model_name + '.sav'
if sent_cnt % 1000 == 0:
print(sent_cnt, 'sentences on', len(formatted_corpus), flush=True)
stack = []
queue = list(sentence)
graph = {}
graph['heads'] = {}
graph['heads']['0'] = '0'
graph['deprels'] = {}
graph['deprels']['0'] = 'ROOT'
transitions = []
x_templist = []
y_templist = []
while queue:
current_dictX, current_Y = features.extract(
stack, queue, graph, feature_1, sentence, 1)
stack, queue, graph, trans = reference(stack, queue, graph)
transitions.append(trans)
x_templist.append(current_dictX)
y_templist.append(current_Y)
stack, graph = transition.empty_stack(stack, graph)
for word in sentence:
word['head'] = graph['heads'][word['id']]
x_list.extend(x_temp_list)
y_list.extend(y_temp_list)
print("Encoding the features and classes...")
def build_instance(tagged_sentence, annotated, index):
feat = features.extract(tagged_sentence, annotated, index)
label = tagged_sentence[index][1]
return (feat, label)
def process(
file
): # used by multiprocessor. Tensors are stored multiple times to reduce RAM usage.
global gamecounter, kgsdatain, kgsdataout, kgsdatawin
print file
gamefile = open(file, 'r')
gamecounter += 1
if gamecounter == SAVEFILE:
lock.acquire()
newin = np.load(
'/media/falk/6,0 TB Volume/19 badukmovies/kgsdatain_alllayer.npy')
newout = np.load(
'/media/falk/6,0 TB Volume/19 badukmovies/kgsdataout_alllayer.npy')
newwin = np.load(
'/media/falk/6,0 TB Volume/19 badukmovies/kgsdatawin_alllayer.npy')
newin = np.concatenate([newin, kgsdatain]) if newin.size else kgsdatain
newout = np.concatenate([newout, kgsdataout
]) if newout.size else kgsdataout
newwin = np.concatenate([newwin, kgsdatawin
]) if newwin.size else kgsdatawin
kgsdatain = np.array([])
kgsdataout = np.array([])
kgsdatawin = np.array([])
print("Spielblock gespeichert")
if len(newin) >= SAVEBIGFILE:
metacounter = 0
while os.path.isfile(
'/media/falk/6,0 TB Volume/19 badukmovies/pro/kgsdatain_alllayer_'
+ str(metacounter) + '.npy'):
metacounter += 1
newinstr = '/media/falk/6,0 TB Volume/19 badukmovies/pro/kgsdatain_alllayer_' + str(
metacounter) + '.npy'
newoutstr = '/media/falk/6,0 TB Volume/19 badukmovies/pro/kgsdataout_alllayer_' + str(
metacounter) + '.npy'
newwinstr = '/media/falk/6,0 TB Volume/19 badukmovies/pro/kgsdatawin_alllayer_' + str(
metacounter) + '.npy'
# this is the final storage location
np.save(newinstr, newin)
np.save(newoutstr, newout)
np.save(newwinstr, newwin)
newout = np.array([])
newin = np.array([])
newwin = np.array([])
print(
"--------------- META Spielblock gespeichert--------------------"
)
np.save(
'/media/falk/6,0 TB Volume/19 badukmovies/kgsdatain_alllayer.npy',
newin)
np.save(
'/media/falk/6,0 TB Volume/19 badukmovies/kgsdataout_alllayer.npy',
newout)
np.save(
'/media/falk/6,0 TB Volume/19 badukmovies/kgsdatawin_alllayer.npy',
newwin)
lock.release()
gamecounter = 0
gamefile = gamefile.read()
gamefile = gamefile.replace("\n", "")
gamefile = gamefile.replace("(;", "")
game = gamefile
boardpos = initboardpos.copy()
if "HA[" in gamefile: # if Handycap
handycap = int(gamefile.split("HA[", 1)[1][0]) #write down Handycap
if handycap >= 2:
(boardpos[0, 15, 3], boardpos[0, 3, 15]) = (1, 1)
(boardpos[2, 15, 3], boardpos[2, 3, 15]) = (0, 0)
if handycap >= 3:
boardpos[0, 15, 15] = 1
boardpos[2, 15, 15] = 0
if handycap >= 4:
boardpos[0, 3, 3] = 1
boardpos[2, 3, 3] = 0
if handycap == 5:
boardpos[0, 9, 9] = 1
boardpos[2, 9, 9] = 0
if handycap >= 6:
(boardpos[0, 9, 3], boardpos[0, 9, 15]) = (1, 1)
(boardpos[2, 9, 3], boardpos[0, 9, 15]) = (0, 0)
if handycap == 7:
boardpos[0, 9, 9] = 1
boardpos[2, 9, 9] = 0
if handycap >= 8:
(boardpos[2, 3, 9], boardpos[0, 15, 9]) = (0, 0)
if handycap == 9:
boardpos[0, 9, 9] = 1
boardpos[2, 9, 9] = 0
# main sgf disassembly begins here
for sig, col, y, x in zip(game, game[1:], game[3:],
game[4:]): # structure: ;B[ic]
######################## store prev board, feedback ##########################
if sig == ";":
if x not in alphabet: break # error or pass: leave game
if y not in alphabet: break # error or pass: leave game
if col not in ("B", "W"): break # error: leave game
xpos = alphabet.index(x)
ypos = alphabet.index(y)
storeboardpos = boardpos[None, :]
kgsdatain = np.concatenate([kgsdatain, storeboardpos
]) if kgsdatain.size else storeboardpos
# store to database (uses boardpos from last iteration)
boardpos2 = boardpos[0:2].copy(
) # swap board for agent to always have color of plane 1
boardpos2[1] = boardpos[0] # 1st agent is b, 2nd w, 3rd b etc.
boardpos[0] = boardpos[1]
boardpos[1] = boardpos2[1]
onemove = np.zeros([1, BOARD, BOARD],
dtype=int) # add one in feedback board
onemove[0, xpos, ypos] = 1
onewin = np.array([1])
if "RE[B" in game: # b wins = 1
onewin[0] = 1
else:
onewin[0] = 0
kgsdatawin = np.concatenate(
[kgsdatawin, onewin]) if kgsdatawin.size else onewin # we lose
kgsdataout = np.concatenate([kgsdataout, onemove
]) if kgsdataout.size else onemove
############################# add new move #################################
boardpos = features.extract(boardpos, xpos, ypos,
col) # modifies all layers in boardpos
def testLetterTrigrams(self):
text = ""
results = features.extract(text,[Features.LETTERTRIGRAMS],False)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(0, sum(result.values()))
results = features.extract(text,[Features.LETTERTRIGRAMS],True)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(0, sum(result.values()))
text = "eee"
results = features.extract(text,[Features.LETTERTRIGRAMS],False)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(1, result['eee'])
self.assertEqual(1, sum(result.values()))
results = features.extract(text,[Features.LETTERTRIGRAMS],True)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(1.0, result['eee'])
self.assertEqual(1.0, sum(result.values()))
text = "eee eee e ee"
results = features.extract(text,[Features.LETTERTRIGRAMS],False)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(2, result['eee'])
self.assertEqual(2, sum(result.values()))
results = features.extract(text,[Features.LETTERTRIGRAMS],True)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(1.0, result['eee'])
self.assertEqual(1.0, sum(result.values()))
text = "eekke eee e ezee zzkzkz"
results = features.extract(text,[Features.LETTERTRIGRAMS],False)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(1, result['eek'])
self.assertEqual(1, result['ekk'])
self.assertEqual(1, result['kke'])
self.assertEqual(1, result['eee'])
self.assertEqual(1, result['eze'])
self.assertEqual(1, result['zee'])
self.assertEqual(1, result['zzk'])
self.assertEqual(2, result['zkz'])
self.assertEqual(1, result['kzk'])
self.assertEqual(10, sum(result.values()))
results = features.extract(text,[Features.LETTERTRIGRAMS],True)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(1/10.0, result['eek'])
self.assertEqual(1/10.0, result['ekk'])
self.assertEqual(1/10.0, result['kke'])
self.assertEqual(1/10.0, result['eee'])
self.assertEqual(1/10.0, result['eze'])
self.assertEqual(1/10.0, result['zee'])
self.assertEqual(1/10.0, result['zzk'])
self.assertEqual(2/10.0, result['zkz'])
self.assertEqual(1/10.0, result['kzk'])
self.assertEqual(10/10.0, round(sum(result.values())))
text = "ababababa"
results = features.extract(text,[Features.LETTERTRIGRAMS],False)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(4, result['aba'])
self.assertEqual(3, result['bab'])
self.assertEqual(7, sum(result.values()))
results = features.extract(text,[Features.LETTERTRIGRAMS],True)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(4/7.0, result['aba'])
self.assertEqual(3/7.0, result['bab'])
self.assertEqual(1.0, sum(result.values()))
text = "ab \nn \ttzx"
results = features.extract(text, [Features.LETTERTRIGRAMS], False)
result = results[Features.LETTERTRIGRAMS]
self.assertEqual(1.0, result['tzx'])
self.assertEqual(1.0, sum(result.values()))
