def makeWekaOptions(learning_rate, momentum, number_hidden, num_cv, costs = None):
"Return Weka option string for specified values"
options_map = {'M':momentum, 'L':learning_rate, 'H':number_hidden, 'x':num_cv}
if costs:
cost_matrix_path = csv.makeTempPath('cost') + '.cost'
options_map['m'] = cost_matrix_path
cost_matrix = ['%% Rows Columns',
spaceSeparatedLine([2,2]),
'%% Matrix elements',
spaceSeparatedLine([0.0, costs['True']]),
spaceSeparatedLine([costs['False'], 0.0])]
file(cost_matrix_path, 'w').write('\n'.join(cost_matrix))
return mapToWekaOptions(options_map)
def testMatrixMLP(matrix, columns, opts = mlp_opts):
"Run MLP on attributes with index in columns"
c_x = columns + [-1] # include outcome
sub_matrix = [[row[i] for i in c_x] for row in matrix]
temp_base = csv.makeTempPath('subset'+('%03d'%len(columns))+'_')
temp_csv = temp_base + '.csv'
temp_results = temp_base + '.results'
if is_testing:
num_attributes = len(matrix[0]) - 1
accuracy,dt = 1.0/float(sum([abs(x-num_attributes/2) for x in columns])), 0.1
else:
csv.writeCsv(temp_csv, sub_matrix)
accuracy,dt = runMLPTrain(temp_csv, temp_results, opts)
return (accuracy, temp_csv, temp_results, dt)
if False:
selectAttibutesGA()
if False:
num_subset = 25
in_filename = csv.makeCsvPath('subset.best' + ('%03d'%num_subset))
csv_results_name = csv.makePath('hidden.layer.results')
csv_summary_name = csv.makeCsvPath('hidden.layer.summary')
csv_best_name = csv.makeCsvPath('hidden.layer.best')
csv_summary = file(csv_summary_name, 'w')
best_accuracy = 0.0
for num_hidden in range(1, num_subset):
opts = '-H ' + str(num_hidden) + ' -x 4'
out_filename = csv.makeCsvPath('num.hidden' + ('%03d'%num_hidden))
temp_base = csv.makeTempPath('num.hidden' + ('%03d'%num_hidden))
temp_results = temp_base + '.results'
accuracy, duration = runMLPTrain(in_filename, temp_results, opts)
summary = [num_hidden, accuracy, best_accuracy, duration, temp_results]
print summary
csv_line = ','.join([str(e) for e in summary])
csv_summary.write(csv_line + '\n')
csv_summary.flush()
if accuracy > best_accuracy:
best_accuracy = accuracy
shutil.copyfile(temp_results, csv_results_name)
if False:
num_subset = 25
in_filename = csv.makeCsvPath('subset.best' + ('%03d'%num_subset))
csv_results_name = csv.makePath('learning.rate.results')