# tf.global_variables_initializer().run() # 初始化
# summary_writer.add_graph(sess.graph) # 写入变量图
ber_wrong_number = []
# saver = tf.train.Saver(max_to_keep=4)
saver = tf.train.import_meta_graph('t/modulate-model-900.meta')
x = tf.get_default_graph().get_operation_by_name('x_input').outputs[0]
final_value = tf.get_default_graph().get_operation_by_name(
'final_value').outputs[0]
y_ber = tf.get_default_graph().get_operation_by_name(
'y-input-for-ber').outputs[0]
ber = tf.reduce_mean(tf.abs(final_value - y_ber))
for i in SNR:
Y = generate_data.generate_bit([TRAIN_NUM,
OUTPUT_NODE]) # 让数据的0和1的概率都相同
X = encode.encode2d(Y) # TRAIN_NUM , OUTPUT_NODE / 2 , 2
Y_one_hot = encode.encode2d_onehot(Y) # 单热码用于验证数据
E_x = 10**(0.1 * i) # 信号能量
noise = np.random.randn(TRAIN_NUM, INPUT_NODE, 2) # sigma * r + mu
X = X * E_x + noise
# saver=tf.train.import_meta_graph('t/modulate-model-900.meta')
saver.restore(sess, tf.train.latest_checkpoint('t/'))
ber_loss = sess.run(ber, feed_dict={x: X, y_ber: Y})
ber_wrong_number.append(ber_loss)
all_wrong = np.sum(ber_wrong_number, axis=0) #axis=1 是按行求和
print('ber为%f' % (ber_loss))
# saver.save(sess,'t/modulate-model',global_step=i)
LEARNING_RATE_BASE = 0.01 # 模型基础学习速率
LEARNING_RATE_DECAY = 0.99 # 学习衰减速度
BATCH_SIZE = 200 # 一批数据量
TRAIN_NUM = 20000 # 数据总量
MOVING_AVERAGE_DECAY = 0.99 # 滑动平均衰减
TRAINING_STEPS = 500 # 训练多少次
SNR = 5 # 信噪比
E_x = 10**(0.1 * SNR) # 信号能量
# 生成数据
Y = np.random.randint(0, 2, [TRAIN_NUM, OUTPUT_NODE]).astype('float32')
X = encode.encode2d(Y) # TRAIN_NUM , OUTPUT_NODE / 2 , 2
# 验证数据
Y_vaildate = np.random.randint(0, 2,
[TRAIN_NUM, OUTPUT_NODE]).astype('float32')
X_validate = encode.encode2d(Y) # TRAIN_NUM , OUTPUT_NODE / 2 , 2
# 定义整个模型的x和y
x = tf.placeholder(tf.float32, [None, INPUT_NODE, 2], name='x_input')
y_ = tf.placeholder(tf.float32, [None, OUTPUT_NODE], name='y-input')
"""
# 发射机通过DNN输出的数据
# dnn_interface(input_tensor, output_shape, regularizer_rate=None, drop=None)
sent_data = DNN_interface.dnn_interface(input_tensor=x,
LEARNING_RATE_BASE = 0.01 # 模型基础学习速率
LEARNING_RATE_DECAY = 0.99 # 学习衰减速度
BATCH_SIZE = 100 # 一批数据量
TRAIN_NUM = 2000 # 数据总量
MOVING_AVERAGE_DECAY = 0.99 # 滑动平均衰减
TRAINING_STEPS = 1000 # 训练多少次
SNR = -2 # 信噪比
E_x = 10**(0.1 * SNR) # 信号能量
# 生成数据
Y = generate_data.generate_bit([TRAIN_NUM, OUTPUT_NODE]) # 让数据的0和1的概率都相同
X = encode.encode2d(Y).reshape(-1,
INPUT_NODE) # TRAIN_NUM , OUTPUT_NODE / 2 , 2
X_onehot = encode.encode2d_onehot(Y)
# 验证数据
Y_vaildate = generate_data.generate_bit([TRAIN_NUM, OUTPUT_NODE])
X_validate = encode.encode2d(Y).reshape(
-1, INPUT_NODE) # TRAIN_NUM , OUTPUT_NODE / 2 , 2
X_validate_onehot = encode.encode2d_onehot(Y)
# 加噪声
X_shape = X.shape
noise = np.random.randn(X_shape[0], X_shape[1]) # sigma * r + mu
X = X * E_x + noise