with open(training_dataset, 'r') as ff:
training_data = pickle.load(ff)
except IOError:
training_data = dict()
bitmap = list()
for period in xrange(dataset_end, dataset_begin - 1, -1):
bit = [False] * 80
for num in all_data[period]['numbers']:
bit[num - 1] = True
bitmap.append(tuple(bit))
bitmap = tuple(bitmap)
for y in xrange(len(bitmap) - 1 - height + 1, -1, -1):
period = dataset_end - y
if period not in training_data:
if period - 1 in training_data:
training_data[period] = copy.deepcopy(training_data[period - 1])
else:
training_data[period] = dict()
for x in xrange(80 - 1 - width + 1, -1, -1):
m = Matrix.create_from_bitmap(width, height, bitmap, x, y)
if m in training_data[period]:
training_data[period][m] += 1
else:
training_data[period][m] = 1
with open(training_dataset, 'w') as ff:
pickle.dump(training_data, ff, pickle.HIGHEST_PROTOCOL)
def predict(self, target_period=None, dataset_begin=None, dataset_end=None, debug=False):
if target_period is None:
target_period = self.get_max_period() + 1
if dataset_begin is None:
dataset_begin = self.get_min_period()
if dataset_end is None:
dataset_end = self.get_max_period()
if dataset_begin > dataset_end:
raise ValueError()
training_dataset = os.path.join(COOKED_DATA_DIR, 'matrix_predictor_%d_%d' % (self._width, self._height))
try:
with open(training_dataset, 'r') as ff:
training_data = pickle.load(ff)
except IOError:
training_data = dict()
if target_period - 1 not in training_data:
bitmap = list()
for period in xrange(dataset_end, dataset_begin - 1, -1):
bit = [False] * 80
for num in self._all_data[period]['numbers']:
bit[num - 1] = True
bitmap.append(tuple(bit))
bitmap = tuple(bitmap)
for y in xrange(len(bitmap) - 1 - self._height + 1, -1, -1):
period = dataset_end - y
if period not in training_data:
if period - 1 in training_data:
training_data[period] = copy.deepcopy(training_data[period - 1])
else:
training_data[period] = dict()
for x in xrange(80 - 1 - self._width + 1, -1, -1):
m = Matrix.create_from_bitmap(self._width, self._height, bitmap, x, y)
if m in training_data[period]:
training_data[period][m] += 1
else:
training_data[period][m] = 1
with open(training_dataset, 'w') as ff:
pickle.dump(training_data, ff, pickle.HIGHEST_PROTOCOL)
sorted_matrices = sorted(training_data[target_period - 1].items(), key=itemgetter(1), reverse=True)
latest_bitmap = list()
for period in xrange(target_period - 1, target_period - 1 - self._height - 1, -1):
bm = [False] * 80
for num in self._all_data[period]['numbers']:
bm[num - 1] = True
latest_bitmap.append(tuple(bm))
latest_bitmap = tuple(latest_bitmap)
candidate = list()
adopted_matrix = list()
adopted_sub_matrix = set()
for m,f in sorted_matrices:
sub_matrix = Matrix.create_sub_matrix(m)
if m.is_qualified() and sub_matrix not in adopted_sub_matrix:
candidate += m.findall(latest_bitmap)
candidate = list(set(candidate))
adopted_matrix.append(m)
adopted_sub_matrix.add(sub_matrix)
if len(candidate) >= self._target:
break
if debug:
output = list()
#rank = 1
#for m in sorted_matrices:
# if m['matrix'].is_qualified():
# output.append('top %d: ' % rank)
# output.append(str(m['matrix']))
# output.append('')
# rank += 1
# if rank == 10:
# break
output.append('adopted matrice:')
for m in adopted_matrix:
output.append(str(m))
output.append('')
output.append('predicted period: %d' % target_period)
output.append('prediction: %s' % sorted(candidate))
else:
output = dict()
output['period'] = '%d' % target_period
output['numbers'] = sorted(candidate)
return output