def inference(inputs, input_units, output_units, is_train=True):
"""
Define the model by model name.
Return:
The logit of the model output.
"""
if FLAGS.model == "dnn":
return model.dnn_inference(inputs, input_units, output_units, is_train,
FLAGS)
elif FLAGS.model == "lr":
return model.lr_inference(inputs, input_units, output_units, is_train,
FLAGS)
elif FLAGS.model == "wide_and_deep":
return model.wide_and_deep_inference(inputs, input_units, output_units,
is_train, FLAGS)
elif FLAGS.model == "customized":
return model.customized_inference(inputs, input_units, output_units,
is_train, FLAGS)
elif FLAGS.model == "cnn":
return model.cnn_inference(inputs, input_units, output_units, is_train,
FLAGS)
elif FLAGS.model == "customized_cnn":
return model.customized_cnn_inference(inputs, input_units, output_units,
is_train, FLAGS)
elif FLAGS.model == "lstm":
return model.lstm_inference(inputs, input_units, output_units, is_train,
FLAGS)
elif FLAGS.model == "bidirectional_lstm":
return model.bidirectional_lstm_inference(inputs, input_units,
output_units, is_train, FLAGS)
elif FLAGS.model == "gru":
return model.gru_inference(inputs, input_units, output_units, is_train,
FLAGS)
def inference(inputs, input_units, output_units, is_train=True):
"""
Define the model by model name.
Return:
The logit of the model output.
"""
if FLAGS.model == "dnn":
return model.dnn_inference(inputs, input_units, output_units, is_train,
FLAGS)
elif FLAGS.model == "lr":
return model.lr_inference(inputs, input_units, output_units, is_train,
FLAGS)
elif FLAGS.model == "wide_and_deep":
return model.wide_and_deep_inference(inputs, input_units, output_units,
is_train, FLAGS)
elif FLAGS.model == "customized":
return model.customized_inference(inputs, input_units, output_units,
is_train, FLAGS)
elif FLAGS.model == "cnn":
return model.cnn_inference(inputs, input_units, output_units, is_train,
FLAGS)
elif FLAGS.model == "customized_cnn":
return model.customized_cnn_inference(inputs, input_units, output_units,
is_train, FLAGS)
elif FLAGS.model == "lstm":
return model.lstm_inference(inputs, input_units, output_units, is_train,
FLAGS)
elif FLAGS.model == "bidirectional_lstm":
return model.bidirectional_lstm_inference(inputs, input_units,
output_units, is_train, FLAGS)
elif FLAGS.model == "gru":
return model.gru_inference(inputs, input_units, output_units, is_train,
FLAGS)
def evaluate_one_image():
'''Test one image against the saved models and parameters
'''
# you need to change the directories to yours.
# train_dir = 'D:/python/deep-learning/CatVsDog/Project/test_image/'
train_dir = 'D:/python/deep-learning/CatVsDog/Project/test_image/'
train, train_label = input_data.get_files(train_dir)
image_array = get_one_image(train)
with tf.Graph().as_default():
BATCH_SIZE = 1
N_CLASSES = 2
image = tf.cast(image_array, tf.float32)
image = tf.image.per_image_standardization(image)
image = tf.reshape(image, [1, 208, 208, 3])
logit = model.cnn_inference(image, BATCH_SIZE, N_CLASSES)
logit = tf.nn.softmax(logit)
x = tf.placeholder(tf.float32, shape=[208, 208, 3])
# you need to change the directories to yours.
logs_train_dir = 'D:/python/deep-learning/CatVsDog/Project/log/'
saver = tf.train.Saver()
with tf.Session() as sess:
print("Reading checkpoints...")
ckpt = tf.train.get_checkpoint_state(logs_train_dir)
if ckpt and ckpt.model_checkpoint_path:
global_step = ckpt.model_checkpoint_path.split('/')[-1].split(
'-')[-1]
saver.restore(sess, ckpt.model_checkpoint_path)
print('Loading success, global_step is %s' % global_step)
else:
print('No checkpoint file found')
prediction = sess.run(logit, feed_dict={x: image_array})
max_index = np.argmax(prediction)
if max_index == 0:
print('This is a cat with possibility %.6f' % prediction[:, 0])
else:
print('This is a dog with possibility %.6f' % prediction[:, 1])
plt.imshow(image_array)
plt.show()
IMG_W = 208 # resize图像，太大的话训练时间久
IMG_H = 208
BATCH_SIZE = 16
CAPACITY = 2000
MAX_STEP = 5000 # 一般5K~10k
learning_rate = 0.0001 # 一般小于0.0001
train_dir = 'F:/Data/train2/'
logs_train_dir = 'F:/Data/doglog/' # 记录训练过程与保存模型
train, train_label = input_data.get_files(train_dir)
train_batch, train_label_batch = input_data.get_batch(train, train_label,
IMG_W, IMG_H, BATCH_SIZE,
CAPACITY)
train_logits = model.cnn_inference(train_batch, BATCH_SIZE, N_CLASSES)
train_loss = model.losses(train_logits, train_label_batch)
train_op = model.training(train_loss, learning_rate)
train__acc = model.evaluation(train_logits, train_label_batch)
summary_op = tf.summary.merge_all() # 这个是log汇总记录
# 可视化为了画折线图
step_list = list(range(100)) # 因为后来的cnn_list加了200个
cnn_list1 = []
cnn_list2 = []
fig = plt.figure() # 建立可视化图像框
ax = fig.add_subplot(1, 1, 1) # 子图总行数、列数，位置
ax.yaxis.grid(True)
ax.set_title('cnn_accuracy ', fontsize=14, y=1.02)
ax.set_xlabel('step')
def train():
tf.reset_default_graph()
sess = tf.Session()
keep_prob_ = tf.placeholder(tf.float32, name='keep')
learning_rate_ = tf.placeholder(tf.float32, name='learning_rate')
inputs, labels, total_count = dataset.csv_inputs(batch_size, epochs,
n_classes, n_channels,
seq_len, trial)
inputs = tf.cast(inputs, tf.float32)
labels = tf.cast(labels, tf.float32)
total_count = tf.cast(total_count, tf.float32)
logits = model.cnn_inference(inputs, keep_prob_, n_classes)
cost = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits_v2(logits=logits,
labels=labels))
tf.summary.scalar("cost", cost)
train_op = tf.train.AdamOptimizer(learning_rate_)
gradients = train_op.compute_gradients(cost)
capped_gradients = [(tf.clip_by_value(grad, -1., 1.), var)
for grad, var in gradients]
optimizer = train_op.apply_gradients(capped_gradients)
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(labels, 1))
lg = tf.argmax(logits, 1)
ll = tf.argmax(labels, 1)
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
tf.summary.scalar("accuracy", accuracy)
summ = tf.summary.merge_all()
saver = tf.train.Saver()
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
#############################################################################
saver.restore(sess, tf.train.latest_checkpoint(save_path))
#############################################################################
writer_train = tf.summary.FileWriter(save_path + 'train_accuracy/',
sess.graph)
print("epoch looping")
index = 0
# Feed dictionary
feed = {keep_prob_: keep_prob, learning_rate_: learning_rate}
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
e = 0
loss_pre = 1000
try:
while not coord.should_stop():
index += 1
logits_val, labels_val, loss, _, acc, s_t, row_count = sess.run(
[lg, ll, cost, optimizer, accuracy, summ, total_count],
feed_dict=feed)
writer_train.add_summary(s_t, index)
train_acc.append(acc)
train_loss.append(loss)
if index % np.floor(row_count / batch_size) == 0:
e += 1
if loss < loss_pre:
saver.save(sess, save_path + 'save.ckpt')
loss_pre = loss
if loss < 0.0000001 and acc == 1:
print("Epoch: {}/{}".format(e, epochs),
"Iteration: {:d}".format(index),
"Train loss: {:.10f}".format(loss),
"Train acc: {:.4f}".format(acc))
saver.save(sess, save_path + 'save.ckpt')
break
# Print at each 1000 iterations
if index % 100 == 0:
print("Epoch: {}/{}".format(e, epochs),
"Iteration: {:d}".format(index),
"Train loss: {:.10f}".format(loss),
"Train acc: {:.4f}".format(acc))
except tf.errors.OutOfRangeError:
print('epoch reached!')
finally:
print("Epoch: {}/{}".format(e,
epochs), "Iteration: {:d}".format(index),
"Final train loss: {:.10f}".format(loss_pre))
coord.request_stop()
coord.join(threads)
sess.close()
def test():
t = time.time()
tf.reset_default_graph()
sess = tf.Session()
keep_prob_ = tf.placeholder(tf.float32, name='keep')
learning_rate_ = tf.placeholder(tf.float32, name='learning_rate')
inputs, labels, trial_num, total_count = dataset.csv_test(
batch_size, n_classes, n_channels, seq_len, trial)
inputs = tf.cast(inputs, tf.float32)
labels = tf.cast(labels, tf.float32)
total_count = tf.cast(total_count, tf.uint16)
logits = model.cnn_inference(inputs, keep_prob_, n_classes)
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(labels, 1))
cost = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits_v2(logits=logits,
labels=labels))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
tf.summary.scalar("accuracy", accuracy)
saver = tf.train.Saver()
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
sess.run(init_op)
saver.restore(sess, tf.train.latest_checkpoint(save_path))
print("epoch looping")
index = 0
feed = {keep_prob_: 1, learning_rate_: 1}
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
elapsed = time.time() - t
t1 = time.time()
try:
while not coord.should_stop():
index += 1
batch_acc, batch_loss, batch_logits, batch_labels, total_counts = sess.run(
[accuracy, cost, logits, labels, total_count], feed_dict=feed)
probabilities.append(
tf.nn.softmax(batch_logits).eval(session=sess))
test_labels.append(tf.argmax(batch_labels, 1).eval(session=sess))
predictions.append(
tf.argmax(tf.nn.softmax(batch_logits), 1).eval(session=sess))
test_acc.append(batch_acc)
test_loss.append(batch_loss)
print(
"Iteration: {}/{}".format(
index,
np.floor(total_counts / batch_size).astype(int)),
"Batch test accuracy: {:.6f}".format(batch_acc))
except tf.errors.OutOfRangeError:
print('epoch reached!')
finally:
coord.request_stop()
coord.join(threads)
elapsed1 = time.time() - t1
clabels = np.concatenate(test_labels, axis=0)
cpredictions = np.concatenate(predictions, axis=0)
cprobabilities = np.concatenate(probabilities, axis=0)
confusion_matrix = tf.confusion_matrix(
labels=clabels, predictions=cpredictions).eval(session=sess)
sess.close()
print("Mean test accuracy: {:.6f}".format(np.mean(test_acc)))
df = pd.DataFrame(confusion_matrix)
df.to_csv(result_path + 'confusion_matrix.csv')
df1 = pd.DataFrame(clabels)
df1.to_csv(result_path + 'labels.csv', header=None, index=None)
df2 = pd.DataFrame(cpredictions)
df2.to_csv(result_path + 'predictions.csv', header=None, index=None)
df3 = pd.DataFrame(test_acc)
df3.to_csv(result_path + 'test_acc.csv', header=None, index=None)
df4 = pd.DataFrame(cprobabilities)
df4.to_csv(result_path + 'probabilities.csv', header=None, index=None)
df5 = pd.DataFrame(test_loss)
df5.to_csv(result_path + 'test_loss.csv', header=None, index=None)
df6 = pd.DataFrame([
'CNN', trial,
np.mean(test_acc),
np.mean(test_loss), elapsed, elapsed1 / index
])
df6.to_csv(result_path + 'result.csv', header=None, index=None)
# rows_sums = confusion_matrix.sum(axis=1)
# normalised_confusion_matrix = confusion_matrix/rows_sums[:, np.newaxis]
print(confusion_matrix)
print(trial)
print('CNN')
print('mean test accuracy: ', np.mean(test_acc))
print('mean test loss: ', np.mean(test_loss))
print('T_model: ', elapsed)
print('T_batch: ', elapsed1 / index)
#一个batch_size，是继续(True)输出还是丢弃(False)。
train_dataset = train_dataset.shuffle(100).batch(BATCH_SIZE, drop_remainder = False).repeat()
#dataset实际也是一个生成器，所以要生成一个iterator来迭代获取其中的数据
train_iterator = tf.data.make_one_shot_iterator(train_dataset)
train_handle = tf.placeholder(tf.string, shape = [])
iterator_from_train_handle = tf.data.Iterator.from_string_handle(
train_handle, train_dataset.output_types, train_dataset.output_shapes)
#关键一步，从迭代器中生成下一批数据
next_train_data, next_train_label = iterator_from_train_handle.get_next()
#定义占位符
x = tf.placeholder(tf.float32, shape = [None, IMAGE_H, IMAGE_W, CHANNELS])
#计算前向传播
y_pred = model.cnn_inference(inputs = x, num_classes = N_CLASSES)
y_true = tf.placeholder(tf.int64, shape = [None])
train_loss = model.losses(logits = y_pred, labels = y_true)
train_accuracy = model.evaluation(logits = y_pred, labels = y_true)
train_op = model.training(train_loss, learning_rate = 0.0001)
#summary_op = tf.summary.merge_all()
init_op = tf.global_variables_initializer()
#定义一个字典，来储存过程中的损失值和准确率
history = {}
history["train_loss"] = []
history["train_acc"] = []