python深度神经网络量化_深度神经网络数据集大小
问题描述
我的数据集才一千多个，是不是用深度神经网络的模型，不够，容易欠拟合
问题出现的环境背景及自己尝试过哪些方法
我之前的训练参照了两层的CIFAR卷积层测试了
用1000次迭代 每次10batch_size
结果
相关代码
// 请把代码文本粘贴到下方(请勿用图片代替代码)
import cv2
import numpy as np
import os
import random
import tensorflow as tf
import sklearn.utils
def read_and_decode(filename, testing = False):
filename_queue = tf.train.string_input_producer([filename])
reader = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
if testing == False:
features = tf.parse_single_example(serialized_example,
features={
'label': tf.FixedLenFeature([], tf.int64),
'img_raw' : tf.FixedLenFeature([], tf.string),
})
img = tf.decode_raw(features['img_raw'], tf.uint8)
img = tf.reshape(img, [600, 328, 1])
img = tf.cast(img, tf.float32) * (1. / 255) - 0.5
label = tf.cast(features['label'], tf.int32)
return img, label
else:
features = tf.parse_single_example(serialized_example,
features={
'label_test': tf.FixedLenFeature([], tf.int64),
'img_raw_test' : tf.FixedLenFeature([], tf.string),
})
img = tf.decode_raw(features['img_raw_test'], tf.uint8)
img = tf.reshape(img, [600, 328, 1])
img = tf.cast(img, tf.float32) * (1. / 255) - 0.5
label = tf.cast(features['label_test'], tf.int32)
return img, label
if name == '__main__':
img, label = read_and_decode("train.tfrecords")
img_train, label_train = tf.train.shuffle_batch([img, label],
batch_size=10, capacity=2000,
min_after_dequeue=1000)
img_raw_test, label_test = read_and_decode("test.tfrecords", testing = True)
img_test, label_test = tf.train.shuffle_batch([img_raw_test, label_test],
batch_size=10, capacity=2000,
min_after_dequeue=1000)
print("begin")
print("begin data")
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev = 0.1)
return tf.Variable(initial)
def bias_variable(shape):
initial = tf.constant(0.1, shape = shape)
return tf.Variable(initial)
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides = [1, 1, 1, 1], padding='SAME')
def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize = [1, 2, 2, 1], strides = [1, 2, 2, 1], padding = 'SAME')
def avg_pool_82x150(x):
return tf.nn.avg_pool(x, ksize = [1, 150, 82, 1], strides = [1, 150, 82, 1], padding = 'SAME')
x = tf.placeholder(tf.float32, [None, 600, 328, 1])
y = tf.placeholder(tf.float32, [None, 6])
W_conv1 = weight_variable([5, 5, 1, 64])
b_conv1 = bias_variable([64])
x_image = tf.reshape(x, [-1, 600, 328, 1])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)
W_conv2 = weight_variable([5, 5, 64, 64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2)
W_conv3 = weight_variable([5, 5, 64, 6])
b_conv3 = bias_variable([6])
h_conv3 = tf.nn.relu(conv2d(h_pool2, W_conv3) + b_conv3)
# 经过两层池化后 图片变成82*150
nt_hpool3 = avg_pool_82x150(h_conv3)
nt_hpool3_flat = tf.reshape( nt_hpool3 , [-1, 6])
y_conv = tf.nn.softmax(nt_hpool3_flat)
cross_entropy = -tf.reduce_sum(y*tf.log(y_conv))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
#开始会话训练
sess = tf.Session()
sess.run(tf.global_variables_initializer())
tf.train.start_queue_runners(sess = sess)
for i in range(1000):
image_batch, label_batch = sess.run([img_train, label_train])
label_b = np.eye(6, dtype =float)[label_batch]
train_step.run(feed_dict = {x:image_batch, y:label_b},session = sess)
if i%20 == 0:
train_accuracy = accuracy.eval(feed_dict = {x:image_batch, y:label_b}, session = sess)
print("step %d, training accuracy %g" %(i, train_accuracy))
image_batch, label_batch = sess.run([img_test, label_test])
label_b = np.eye(6, dtype = float)[label_batch]
print("finished!test accuracy %g" %accuracy.eval(feed_dict = {x: image_batch, y:label_b}, session = sess))
你期待的结果是什么？实际看到的错误信息又是什么？
可以看到 这里的泛化能力还是挺弱的
是不是欠拟合了
还有数据集是不是个硬伤