def run_cnn_with_best_params(smell, input_data, conv_layers, filter, kernel,
pooling_window, epochs):
outfile = get_out_file(smell + "final")
write_result(
outfile,
"conv_layers,filters,kernel,max_pooling_window,epoch,stopped_epoch,auc,accuracy,precision,recall,f1,average_precision,time\n"
)
try:
config = configuration.CNN_config(conv_layers, filter, kernel,
pooling_window, epochs)
start_time = time.time()
auc, accuracy, precision, recall, f1, average_precision, stopped_epoch = \
cnn(input_data, config, smell, is_final=True)
end_time = time.time()
elapsed_time = end_time - start_time
write_result(
outfile,
str(conv_layers) + "," + str(filter) + "," + str(kernel) + "," +
str(pooling_window) + "," + str(epochs) + "," +
str(stopped_epoch) + "," + str(auc) + "," + str(accuracy) + "," +
str(precision) + "," + str(recall) + "," + str(f1) + "," +
str(average_precision) + "," + str(elapsed_time) + "\n")
except ValueError as error:
print(
"Skipping combination layer: {}, filter: {}, kernel: {}, pooling_window: {}"
.format(conv_layers, filter, kernel, pooling_window))
print(error)
write_result(
outfile,
str(conv_layers) + "," + str(filter) + "," + str(kernel) + "," +
str(pooling_window) + "," + str(epochs) +
",-1,-1,-1,-1,-1,-1,-1,-1\n")
def main_cnn(smell, data_path, max_encoding_dim=1024):
input_data = get_all_data(data_path, smell)
filters = [8, 16, 32, 64]
kernels = [5, 7, 11]
pooling_windows = [2, 3, 4, 5]
layers = [1, 2]
epochs = 20
outfile = get_out_file(smell, "cnn")
write_result(outfile, "conv_layers,filters,kernel,max_pooling_window,epochs,precision,recall,f1,time\n")
for layer in layers:
for filter in filters:
for kernel in kernels:
for pooling_window in pooling_windows:
config = configuration.CNN_config(layer, filter, kernel, pooling_window, epochs)
start_time = time.time()
try:
optimal_threshold, max_pr, max_re, max_f1 = autoencoder_cnn(input_data, config)
except ValueError as error:
print(error)
optimal_threshold = -1
max_pr = -1
max_re = -1
max_f1 = -1
end_time = time.time()
time_taken = end_time - start_time
write_result(outfile,
str(layer) + "," + str(filter) + "," + str(
kernel) + "," + str(pooling_window) + "," + str(epochs) + "," +
str(max_pr) + "," + str(max_re) + "," + str(max_f1) + "," + str(time_taken) + "\n")
def main(training_data_path, eval_data_path, smell, skip_iter=-1):
input_data = get_all_data(training_data_path, eval_data_path, smell)
conv_layers = {1, 2, 3}
filters = {8, 16, 32, 64}
kernels = {5, 7, 11}
pooling_windows = {2, 3, 4, 5}
epochs = {50}
total_iterations = len(conv_layers) * len(filters) * len(kernels) * len(
pooling_windows) * len(epochs)
cur_iter = 1
outfile = get_out_file(smell)
write_result(
outfile,
"conv_layers,filters,kernel,max_pooling_window,epoch,stopped_epoch,auc,accuracy,precision,recall,f1,average_precision,time\n"
)
for layer in conv_layers:
for filter in filters:
for kernel in kernels:
for pooling_window in pooling_windows:
for epoch in epochs:
print("** Iteration {0} of {1} **".format(
cur_iter, total_iterations))
if cur_iter < skip_iter:
cur_iter += 1
continue
try:
config = configuration.CNN_config(
layer, filter, kernel, pooling_window, epoch)
start_time = time.time()
auc, accuracy, precision, recall, f1, average_precision, stopped_epoch = rq1_cnn_2d.cnn(
input_data,
config,
smell,
out_folder=OUT_FOLDER,
dim=DIM)
end_time = time.time()
elapsed_time = end_time - start_time
write_result(
outfile,
str(layer) + "," + str(filter) + "," +
str(kernel) + "," + str(pooling_window) + "," +
str(epoch) + "," + str(stopped_epoch) + "," +
str(auc) + "," + str(accuracy) + "," +
str(precision) + "," + str(recall) + "," +
str(f1) + "," + str(average_precision) + "," +
str(elapsed_time) + "\n")
except ValueError as error:
print(
"Skipping combination layer: {}, filter: {}, kernel: {}, pooling_window: {}"
.format(layer, filter, kernel, pooling_window))
print(error)
write_result(
outfile,
str(layer) + "," + str(filter) + "," +
str(kernel) + "," + str(pooling_window) + "," +
str(epoch) + ",-1,-1,-1,-1,-1,-1,-1,-1\n")
cur_iter += 1