分析一套源代码的代码规范和风格并讨论如何改进优化代码

结合工程实践选题相关的一套源代码，根据其编程语言或项目特点，分析其在源代码目录结构、文件名/类名/函数名/变量名等命名、接口定义规范和单元测试组织形式等方面的做法和特点；

使用的代码为手写汉字识别的代码

https://github.com/chongyangtao/DeepHCCR

目录结构为：

images包括汉字图片

meanfiles包括主要的文件

models包括google的lenet模型

util包括各种文本文件，和两种网络的准确率图

目录结构清晰明了，文件名符合命名规范，容易使人知道各个文件的作用。

部分代码如下

#coding=utf-8
import numpy as np
import pickle
import os
import time
import sys
import shutil
import skimage 

caffe_root = '/home/cscl/caffe-master/' 
sys.path.insert(0, caffe_root + 'python')
import caffe

net_file = 'googlenet_deploy.prototxt'
caffe_model = 'models/googlenet_hccr.caffemodel' 
mean_file = 'meanfiles/CASIA1.0_1.1_1.2_mean_112.npy'
unicode_index = np.loadtxt('util/unicode_index.txt', delimiter = ',',dtype = np.int) #7534
net = caffe.Net(net_file,caffe_model,caffe.TEST)


def get_crop_image(imagepath, img_name):    
    img=skimage.io.imread(imagepath + img_name,as_grey=True)
    black_index = np.where(img < 255 )
    min_x = min(black_index[0])
    max_x = max(black_index[0])
    min_y = min(black_index[1])
    max_y = max(black_index[1])
    #print(min_x,max_x,min_y,max_y)
    image = caffe.io.load_image(imagepath+"//"+img_name)
    return image[min_x:max_x, min_y:max_y,:]


def evaluate(imagepath, top_k):
    transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
    transformer.set_transpose('data', (2,0,1))
    transformer.set_raw_scale('data', 255) 

    rightcount=0
    allcount=0

    allimage=os.listdir(imagepath)

    for img_name in allimage:
       allcount = allcount + 1
       label_truth = img_name.split('.')[0]


       print "----------------------"
       image = get_crop_image(imagepath,img_name)
       net.blobs['data'].data[...] = transformer.preprocess('data',image)
       out = net.forward()
       label_index = net.blobs['loss'].data[0].flatten().argsort()[-1:-top_k-1:-1]
       labels = unicode_index[label_index.astype(np.int)]  # output unicode

       #print 'Index: ',label_index 
       print 'Top-' + str(top_k) + ' Label: ',labels
       print 'label_truth: ',label_truth

       for i in range(0,top_k):
            if  labels[i] == int(label_truth):
                rightcount=rightcount+1
                break
    print(rightcount,allcount,(float)(rightcount)/(float)(allcount))   

if __name__=='__main__':
    imagepath='images/'
    top_k = 1;
    evaluate(imagepath,top_k)

函数，变量名等命名采用XX_XX的方式，清晰的表示了各变量和函数的作用。

同时每一部分功能中间用空行隔开，便于区分每一部分功能。

代码注释偏少，可能部分代码较难理解。

总结同类编程语言或项目在代码规范和风格的一般要求。

头部需加 #coding=utf-8

每一行尽可能不超过80个字符。

缩进使用4个空格。

在二元运算符左右各有一个空格。

对于函数名和变量名均使用小写。

#coding=utf-8import numpy as npimport pickleimport osimport timeimport sysimport shutilimport skimage
caffe_root = '/home/cscl/caffe-master/' sys.path.insert(0, caffe_root + 'python')import caffe
net_file = 'googlenet_deploy.prototxt'caffe_model = 'models/googlenet_hccr.caffemodel' mean_file = 'meanfiles/CASIA1.0_1.1_1.2_mean_112.npy'unicode_index = np.loadtxt('util/unicode_index.txt', delimiter = ',',dtype = np.int) #7534net = caffe.Net(net_file,caffe_model,caffe.TEST)

def get_crop_image(imagepath, img_name):img=skimage.io.imread(imagepath + img_name,as_grey=True)black_index = np.where(img < 255 )min_x = min(black_index[0])max_x = max(black_index[0])min_y = min(black_index[1])max_y = max(black_index[1])#print(min_x,max_x,min_y,max_y)image = caffe.io.load_image(imagepath+"//"+img_name)return image[min_x:max_x, min_y:max_y,:]

def evaluate(imagepath, top_k):transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})transformer.set_transpose('data', (2,0,1))transformer.set_raw_scale('data', 255)
rightcount=0allcount=0
allimage=os.listdir(imagepath)
for img_name in allimage: allcount = allcount + 1 label_truth = img_name.split('.')[0]

print "----------------------" image = get_crop_image(imagepath,img_name) net.blobs['data'].data[...] = transformer.preprocess('data',image) out = net.forward() label_index = net.blobs['loss'].data[0].flatten().argsort()[-1:-top_k-1:-1] labels = unicode_index[label_index.astype(np.int)] # output unicode
#print 'Index: ',label_index print 'Top-' + str(top_k) + ' Label: ',labels print 'label_truth: ',label_truth
for i in range(0,top_k):if labels[i] == int(label_truth):rightcount=rightcount+1breakprint(rightcount,allcount,(float)(rightcount)/(float)(allcount))
if __name__=='__main__':imagepath='images/'top_k = 1;evaluate(imagepath,top_k)

来源：https://www.cnblogs.com/lclclclc/p/11660752.html

标签

min函数