def get_mean_exposures_per_pivot():
"Returns the number of exposures each pivot received."
cursor = connection.cursor()
cursor.execute("""
SELECT pivot, pivot_type, COUNT(*) as n_exposures
FROM drill_question
GROUP BY CONCAT(pivot, "|", pivot_type)
""")
data = cursor.fetchall()
word_c = []
kanji_c = []
combined_c = []
kanji_inc_dist = FreqDist()
for pivot, pivot_type, n_exposures in data:
combined_c.append(n_exposures)
if pivot_type == 'k':
kanji_c.append(n_exposures)
kanji_inc_dist.inc(pivot, n_exposures)
elif pivot_type == 'w':
word_c.append(n_exposures)
for kanji in scripts.unique_kanji(pivot):
kanji_inc_dist.inc(kanji, n_exposures)
else:
raise ValueError('unknown pivot type: %s' % pivot_type)
return [
('Words', mean(word_c)),
('Kanji', mean(kanji_c)),
('Combined', mean(combined_c)),
('Kanji combined', mean(kanji_inc_dist.values())),
]
def get_mean_exposures_per_pivot():
"Returns the number of exposures each pivot received."
cursor = connection.cursor()
cursor.execute("""
SELECT pivot, pivot_type, COUNT(*) as n_exposures
FROM drill_question
GROUP BY CONCAT(pivot, "|", pivot_type)
""")
data = cursor.fetchall()
word_c = []
kanji_c = []
combined_c = []
kanji_inc_dist = FreqDist()
for pivot, pivot_type, n_exposures in data:
combined_c.append(n_exposures)
if pivot_type == 'k':
kanji_c.append(n_exposures)
kanji_inc_dist.inc(pivot, n_exposures)
elif pivot_type == 'w':
word_c.append(n_exposures)
for kanji in scripts.unique_kanji(pivot):
kanji_inc_dist.inc(kanji, n_exposures)
else:
raise ValueError('unknown pivot type: %s' % pivot_type)
return [
('Words', mean(word_c)),
('Kanji', mean(kanji_c)),
('Combined', mean(combined_c)),
('Kanji combined', mean(kanji_inc_dist.values())),
]
def _embellish(response_data):
"""Adds kanji contained in words as kanji exposed."""
kanji_script = scripts.Script.Kanji
for pivot, pivot_type, is_correct, timestamp in response_data:
yield (pivot, pivot_type, is_correct, timestamp)
if pivot_type == 'w' and scripts.contains_script(kanji_script, pivot):
for kanji in scripts.unique_kanji(pivot):
yield kanji, 'k', is_correct, timestamp
def validate(cls):
for syllabus in cls.objects.all():
kanji_set = set(k.kanji for k in \
lexicon_models.Kanji.objects.filter(
partialkanji__syllabus=syllabus)
)
for partial_lexeme in syllabus.partiallexeme_set.all():
for lexeme_surface in partial_lexeme.surface_set.all():
if not scripts.unique_kanji(
lexeme_surface.surface).issubset(kanji_set):
raise Exception('invalid surface')
def validate(cls):
for syllabus in cls.objects.all():
kanji_set = set(k.kanji for k in \
lexicon_models.Kanji.objects.filter(
partialkanji__syllabus=syllabus)
)
for partial_lexeme in syllabus.partiallexeme_set.all():
for lexeme_surface in partial_lexeme.surface_set.all():
if not scripts.unique_kanji(lexeme_surface.surface
).issubset(kanji_set):
raise Exception('invalid surface')
def create_kanji_graph(word_dict=get_word_freq_dict()):
kgraph = nx.Graph()
#word_dict = get_word_freq_dict()
for rank, word in enumerate(get_common_word_list(), start=1):
charpairs = permutations(
scripts.unique_kanji(word_dict[word]['chars']), 2)
wordweight = get_word_importance_by_rank(rank)
for char1, char2 in charpairs:
try:
kgraph[char1][char2]['words'].append((word, rank))
word_list = sorted(set(kgraph[char1][char2]['words']))
weight = kgraph[char1][char2]['weight'] - wordweight
kgraph.add_edge(char1, char2, words=word_list, weight=weight)
except KeyError:
kgraph.add_edge(char1,
char2,
words=[(word, rank)],
weight=-wordweight)
os.chdir(BASE_PATH)
with open('output/kanji_graph_data.txt', 'w', encoding='utf-8') as outfile:
outfile.write("#Kanji Graph Data\n\n## DEGREE:\n\n")
outfile.write(
pp.pformat(sorted(kgraph.degree, key=lambda x: x[1],
reverse=True)))
outfile.write("\n\n##CONNECTED COMPONENTS:\n\n")
outfile.write(
pp.pformat(
sorted(nx.connected_components(kgraph), key=len,
reverse=True)))
outfile.write("\n\n##CLUSTERING:\n\n")
outfile.write(pp.pformat(sorted(nx.clustering(kgraph), reverse=True)))
outfile.write("\n\n##CENTRALITY:\n\n")
outfile.write(
pp.pformat(sorted(nx.betweenness_centrality(kgraph),
reverse=True)))
outfile.write("\n\n##EDGES:\n\n")
outfile.write(
pp.pformat(
sorted(kgraph.edges(data=True),
key=lambda x: x[1],
reverse=True)))
outfile.write("\n\n##MINIMUM SPANNING TREE:\n\n")
outfile.write(
pp.pformat(
sorted(nx.minimum_spanning_tree(kgraph).edges(data=True))))
print('Wrote graph data to ' + BASE_PATH +
'output/kanji_graph_data.txt')
return kgraph
def update_kanji_db():
for rank, word in enumerate(get_common_word_list(), start=1):
for char in scripts.unique_kanji(word):
if rank <= 500:
db.kanji.update({'char': char},
{'$addToSet': {
'top500words': word
}},
upsert=True)
elif rank <= 1000:
db.kanji.update({'char': char},
{'$addToSet': {
'top1000words': word
}},
upsert=True)
elif rank <= 3000:
db.kanji.update({'char': char},
{'$addToSet': {
'top3000words': word
}},
upsert=True)
elif rank <= 5000:
db.kanji.update({'char': char},
{'$addToSet': {
'top5000words': word
}},
upsert=True)
elif rank <= 10000:
db.kanji.update({'char': char},
{'$addToSet': {
'top10000words': word
}},
upsert=True)
elif rank <= 15000:
db.kanji.update({'char': char},
{'$addToSet': {
'top15000words': word
}},
upsert=True)
elif rank <= 20000:
db.kanji.update({'char': char},
{'$addToSet': {
'top20000words': word
}},
upsert=True)
else:
break
def _store_word_surfaces(syllabus, syllabus_bundle):
"""
Aligns the word surfaces with JMdict. We also use our known kanji set,
replacing any unknown kanji with their readings. If this results in a
surface known to JMdict, then we add that surface to the lexeme's list.
"""
_log.start('Building lexeme surfaces', nSteps=2)
_store_reduced_surfaces(syllabus, syllabus_bundle)
_log.log('Adding non-syllabus surfaces which match')
for partial_lexeme in syllabus.partiallexeme_set.all():
# Only add new surfaces if we had no matches
if partial_lexeme.surface_set.count() > 0:
continue
for lexeme_surface in partial_lexeme.lexeme.surface_set.all():
if scripts.unique_kanji(lexeme_surface.surface).issubset(
syllabus_bundle.chars):
partial_lexeme.surface_set.add(lexeme_surface)
_log.finish()
def _store_reduced_surfaces(syllabus, syllabus_bundle):
"""
We may have some but not all of the kanji found in a surface available
as part of the syllabus. In these cases, we see if variants of the surface
which don't use the missing kanji are also available.
"""
_log.start('Finding reduced surfaces', nSteps=1)
n_aligned = 0
for alignment in syllabus_bundle.alignments:
if not alignment.has_kanji():
continue
# Find the word which this alignment represents.
try:
partial_lexeme = syllabus.partiallexeme_set.filter(
lexeme__reading_set__reading=alignment.phoneme).get(
lexeme__surface_set__surface=alignment.grapheme)
except:
continue
reading = partial_lexeme.reading_set.get(reading=alignment.phoneme)
surface = partial_lexeme.lexeme.surface_set.get(
surface=alignment.grapheme)
if scripts.unique_kanji(surface.surface).issubset(
syllabus_bundle.chars):
partial_lexeme.surface_set.add(surface)
syllabus.alignment_set.get_or_create(
reading=reading,
surface=surface,
alignment=alignment.short_form(),
)
n_aligned += 1
_log.finish('%d reduced surfaces' % n_aligned)
return
def get_list(list_name):
"Returns the kanji in the given list."
f = cjkdata.get_resource('lists/char/%s' % list_name)
with codecs.open(f, 'r', 'utf8') as istream:
return scripts.unique_kanji(istream.read())
def get_list(list_name):
"""Returns the kanji in the given list."""
f = cjkdata.get_resource('lists/char/%s' % list_name)
with codecs.open(f, 'r', 'utf8') as istream:
return scripts.unique_kanji(istream.read())
