def fuzzify(membership, row_scores, column_scores, num_iterations, iteration_result,
add_fuzz):
"""Provide an iteration-specific fuzzification"""
if add_fuzz == 'none':
logging.debug('DO NOT FUZZIFY !!')
return row_scores, column_scores
# truth table maps from add_fuzz parameter to where fuzz should be added
fuzz_vals = {'both': (True, True), 'rows': (True, False), 'columns': (False, True)}
fuzz_rows, fuzz_cols = fuzz_vals[add_fuzz]
iteration = iteration_result['iteration']
#logging.debug("__fuzzify(), setup...")
#start_time = util.current_millis()
fuzzy_coeff = old_fuzzy_coefficient(iteration, num_iterations)
iteration_result['fuzzy-coeff'] = fuzzy_coeff
if fuzz_rows:
num_row_fuzzy_values = row_scores.num_rows * row_scores.num_columns
row_score_values = row_scores.values
row_sd_values = []
# iterate the row names directly
row_names = row_scores.row_names
for col in xrange(row_scores.num_columns):
cluster_rows = membership.rows_for_cluster(col + 1)
for row in xrange(row_scores.num_rows):
if row_names[row] in cluster_rows:
row_sd_values.append(row_score_values[row, col])
# Note: If there are no non-NaN values in row_sd_values, row_rnorm
# will have all NaNs
row_rnorm = util.sd_rnorm(row_sd_values, num_row_fuzzy_values, fuzzy_coeff)
row_score_values += np.array(row_rnorm).reshape(row_scores.num_rows,
row_scores.num_columns)
if fuzz_cols:
num_col_fuzzy_values = column_scores.num_rows * column_scores.num_columns
col_score_values = column_scores.values
col_sd_values = []
row_names = column_scores.row_names
for col in xrange(column_scores.num_columns):
cluster_cols = membership.columns_for_cluster(col + 1)
for row in xrange(column_scores.num_rows):
if row_names[row] in cluster_cols:
col_sd_values.append(col_score_values[row, col])
# Note: If there are no non-NaN values in col_sd_values, col_rnorm
# will have all NaNs
col_rnorm = util.sd_rnorm(col_sd_values, num_col_fuzzy_values, fuzzy_coeff)
# add fuzzy values to the row/column scores
col_score_values += np.array(col_rnorm).reshape(column_scores.num_rows,
column_scores.num_columns)
#elapsed = util.current_millis() - start_time
#logging.debug("fuzzify() finished in %f s.", elapsed / 1000.0)
return row_scores, column_scores
def test_sd_rnorm(self):
result = util.sd_rnorm([1.3, 1.6, 1.2, 1.05], 9, 0.748951)
# the results are fairly random, make sure we have the right
# number of values
self.assertEquals(9, len(result))
def fuzzify(membership, row_scores, column_scores, num_iterations,
iteration_result, add_fuzz):
"""Provide an iteration-specific fuzzification"""
if add_fuzz == 'none':
logging.debug('DO NOT FUZZIFY !!')
return row_scores, column_scores
# truth table maps from add_fuzz parameter to where fuzz should be added
fuzz_vals = {
'both': (True, True),
'rows': (True, False),
'columns': (False, True)
}
fuzz_rows, fuzz_cols = fuzz_vals[add_fuzz]
iteration = iteration_result['iteration']
#logging.debug("__fuzzify(), setup...")
#start_time = util.current_millis()
fuzzy_coeff = old_fuzzy_coefficient(iteration, num_iterations)
iteration_result['fuzzy-coeff'] = fuzzy_coeff
if fuzz_rows:
num_row_fuzzy_values = row_scores.num_rows * row_scores.num_columns
row_score_values = row_scores.values
row_sd_values = []
# iterate the row names directly
row_names = row_scores.row_names
for col in xrange(row_scores.num_columns):
cluster_rows = membership.rows_for_cluster(col + 1)
for row in xrange(row_scores.num_rows):
if row_names[row] in cluster_rows:
row_sd_values.append(row_score_values[row, col])
# Note: If there are no non-NaN values in row_sd_values, row_rnorm
# will have all NaNs
row_rnorm = util.sd_rnorm(row_sd_values, num_row_fuzzy_values,
fuzzy_coeff)
row_score_values += np.array(row_rnorm).reshape(
row_scores.num_rows, row_scores.num_columns)
if fuzz_cols:
num_col_fuzzy_values = column_scores.num_rows * column_scores.num_columns
col_score_values = column_scores.values
col_sd_values = []
row_names = column_scores.row_names
for col in xrange(column_scores.num_columns):
cluster_cols = membership.columns_for_cluster(col + 1)
for row in xrange(column_scores.num_rows):
if row_names[row] in cluster_cols:
col_sd_values.append(col_score_values[row, col])
# Note: If there are no non-NaN values in col_sd_values, col_rnorm
# will have all NaNs
col_rnorm = util.sd_rnorm(col_sd_values, num_col_fuzzy_values,
fuzzy_coeff)
# add fuzzy values to the row/column scores
col_score_values += np.array(col_rnorm).reshape(
column_scores.num_rows, column_scores.num_columns)
#elapsed = util.current_millis() - start_time
#logging.debug("fuzzify() finished in %f s.", elapsed / 1000.0)
return row_scores, column_scores
def test_sd_rnorm(self):
result = util.sd_rnorm([1.3, 1.6, 1.2, 1.05], 9, 0.748951)
# the results are fairly random, make sure we have the right
# number of values
self.assertEquals(9, len(result))