app = "_" + str(time.time())
model_file_raw += app
model_file = model_file_raw + ".h5"
# create tensorflow graph session
graph = tf.Graph()
with tf.Session(graph=graph):
model = deep_learning.create_model(x_train, y_train,
config["activation_fn"],
config["loss_fn"])
deep_learning.dl_save_model(model, model_file)
deep_learning.eval_write_info(model, x_eval, y_eval, model_file, 0,
0)
acc = deep_learning.eval_model(model, x_eval, y_eval, sizex[1],
False)
if acc > top_acc:
top_acc = acc
top_model_filename = model_file_raw
if i == n_reps - 1:
if save_best_model:
copy2(top_model_filename + ".h5",
top_model_filename + "_top.h5")
copy2(top_model_filename + ".h5.txt",
top_model_filename + "_top.h5.txt")
# K.clear_session()
y = prep.encode(prep.adapt_input(y))
X = np.array(X)
sizex = np.shape(X)
sizey = np.shape(y)
print("refactored: ")
print(sizex)
print(sizey)
# create tensorflow graph session
tfgraph = tf.Graph()
with tf.Session(graph=tfgraph):
model = deep_learning.dl_load_model(model_file)
acc = deep_learning.eval_model(model, X, y, sizex[1], use_rnn)
# make probability predictions with the model
predictions = deep_learning.predict_model_RNN(model, X)
sizep = np.shape(predictions)
print("prediction shape: ")
print(sizep)
if config["one_hot_encoding"]:
predictions = np.reshape(predictions, (sizep[0], sizep[2]))
sizep = np.shape(predictions)
print("reshape predictions: ")
print(sizep)
predictions = deep_learning.binarize_predictions_max(predictions)
train_percent = config["train_percent"]
train_percent = 0
x_train, y_train = classifiers.split_dataset_train(
x, y, train_percent)
x_eval, y_eval = classifiers.split_dataset_test(
x, y, train_percent)
# acc = deep_learning.eval_model(model, x_train, y_train, sizex[1])
# results_train.append(
# [str(i + 1), str(j + 1), model_file, filename, str(acc*100)])
acc = deep_learning.eval_model(
model, x_eval, y_eval, sizex[1], use_rnn)
results_test.append(
[str(i + 1), str(j + 1), model_file, filename, str(acc*100)])
print("model: " + model_file + "\tdataset: " +
data_file + "\tacc (test):" + str(acc))
# with open(root_data_folder + "/output/" + "cross_check_deep_train.csv", "w") as f:
# for r in results_train:
# f.write(",".join(r) + ",\n")
if use_rnn:
output_file = "cross_check_deep_2_rnn_test.csv"
else:
output_file = "cross_check_deep_1_test.csv"
for i in range(n_reps):
print("evaluating model rep: " + str(i) + "/" + str(n_reps))
# session = K.get_session()
model_file = root_crt_model_folder + "/" + model_filenames[fn]
model_file_raw = model_file
model_file_raw += "_" + str(i+1)
model_file = model_file_raw + ".h5"
# create tensorflow graph session
graph = tf.Graph()
with tf.Session(graph=graph):
model = deep_learning.dl_load_model(model_file)
acc_train = deep_learning.eval_model(
model, x_train, y_train, sizex[1], use_rnn)
dt = time.time()
acc_test = deep_learning.eval_model(
model, x_eval, y_eval, sizex[1], use_rnn)
dt = time.time() - dt
if acc_train > top_train:
top_train = acc_train
if acc_test > top_test:
top_test = acc_test
if dt > top_test_dt:
top_test_dt = dt