def testing_whole_family(outputPath=None, wsChildrenDic=dict(), word2ballDic=dict(), outputBallFile=None):
"""
:param outputPath:
:param wsChildrenDic:
:param word2ballDic:
:param outputBallFile:
:return:
"""
print("checking whether the tree structure is perfectly encoded in nball embeddings...\n")
failed_P, failed_DC = [], []
maxsize, mindim, word2ballDic = load_balls(ipath = outputPath, word2ballDic=word2ballDic)
for froot in get_children('*root*', wsChildrenDic=wsChildrenDic):
failed_P += check_P_for_child_parent_in_one_family(froot,
wsChildrenDic=wsChildrenDic,
word2ballDic=word2ballDic,
ballPath=outputPath)
failed_DC += check_DC_for_sibilings_in_one_family(root='*root*', wsChildrenDic=wsChildrenDic,
word2ballDic=word2ballDic)
print("failed families with P", failed_P)
print("failed families with DC", failed_DC)
if failed_P == [] and failed_DC == []:
print("the tree structure is perfectly encoded in nball embeddings.\n")
print("generating nball embedding file...\n")
merge_balls_into_file(ipath= outputPath, outfile=outputBallFile)
else:
print("the tree structure is NOT perfectly encoded in nball embeddings.\n")
print("try again, or contact the author")
def train_word2ball(root="", outputPath = '', logFile='', wsChildrenDic=dict(),
word2ballDic=dict(), word2vecDic=dict(), outputPathBack = None,
wscatCodeDic=dict(), outputBallFile=None):
"""
:param root:
:param outputPath:
:param logFile:
:param wsChildrenDic:
:param word2ballDic:
:param word2vecDic:
:param wscatCodeDic:
:param outputBallFile:
:param outputBallForestFile:
:return:
"""
training_all_families(root=root, wsChildrenDic=wsChildrenDic, word2vecDic=word2vecDic,
wscatCodeDic=wscatCodeDic, word2ballDic=word2ballDic,
outputPath=outputPath, logFile=logFile)
if outputPathBack:
copy_tree(outputPath, outputPathBack)
maxsize, mindim , word2ballDic = load_balls(ipath=outputPath, word2ballDic=word2ballDic)
fix_dim(maxsize, mindim, bPath=outputPath, outputPath=outputPath)
make_DC_for_first_level_children(root=root, firstChild = 'entity.n.01', wsChildrenDic=wsChildrenDic,
word2ballDic=word2ballDic, outputPath=outputPath,
maxsize=maxsize, mindim=mindim, logFile=logFile)
testing_whole_family(outputPath=outputPath, wsChildrenDic=wsChildrenDic, outputBallFile=outputBallFile)
def training_all_families(root="*root*", wsChildrenDic=dict(), word2vecDic=dict(), wscatCodeDic=dict(),
word2ballDic=dict(),
outputPath=None, logFile=None, checking = False):
"""
:param root:
:param wsChildrenDic:
:param word2vecDic:
:param wscatCodeDic:
:param word2ballDic:
:param outputPath:
:param logFile:
:param checking:
:return:
"""
global L0, DIM
children = get_children(root, wsChildrenDic=wsChildrenDic)
child0= 'entity.n.01'
children = sorted(children, key=lambda ele: np.dot(get_word2vector(child0, word2vecDic=word2vecDic),
get_word2vector(ele, word2vecDic=word2vecDic)))
print(children)
N = int(np.ceil(np.log(len(children))))
open(logFile, 'w+')
while children:
child = children.pop()
k = 512
addDim0 = list(bin(N))[2:][:DIM]
if len(addDim0) < DIM:
addDim0 += [0] * (DIM - len(addDim0))
addDim = [int(ele) * 2 - 1 for ele in addDim0]
addDim = [ele * k for ele in addDim]
print("***", child)
with open(logFile, 'a+') as wlog:
wlog.write(" ".join([str(ele) for ele in [child]
+addDim
+[time.strftime("%Y-%m-%d %H:%M:%S",time.gmtime())]]))
wlog.write("\n")
word2ballDic = training_one_family(root=child, addDim=addDim, wsChildrenDic=wsChildrenDic,
word2vecDic=word2vecDic, wscatCodeDic=wscatCodeDic,
word2ballDic=word2ballDic,
outputPath=outputPath, logFile=logFile)
children = sorted(children, key=lambda ele: np.dot(get_word2vector(child, word2vecDic=word2vecDic),
get_word2vector(ele, word2vecDic=word2vecDic)))
print("finished training of all families\n")
if checking:
print("checking each family\n")
maxsize, mindim, word2ballDic = load_balls(ipath=outputPath, word2ballDic=word2ballDic)
failed_P, failed_DC = [], []
for child in get_children(root):
failed_P += check_P_for_child_parent_in_one_family(child, word2ballDic =word2ballDic,
wsChildrenDic=wsChildrenDic, ballPath=outputPath)
failed_DC += check_DC_for_sibilings_in_one_family(root=child, word2ballDic =word2ballDic,
wsChildrenDic=wsChildrenDic)
print("failed families with P", failed_P)
print("failed families with DC", failed_DC)
return word2ballDic
