Reiase's Blog

最近做Adaboost人脸检测训练，感觉为每一个功能都写一个命令行入口太麻烦，比如，三种训练算法，其Python实现的入口分别是train_1, train_2, train_3。通常需要为每一个训练算法写一个入口脚本，接受命令行参数，完成对训练算法的调用。但是这样太繁琐了，工作目录下一下子多出好多文件，而且要保持训练算法和对应入口脚本的一致性。我编写了一个从shell直接调用python函数的小工具，省去了编写入口脚本的麻烦。比如如下定义的Python函数(foo.py)

def foofunc(arg1, arg2, arg3):
    ...
    return retval

可通过如下命令调用：

$ ./script foofunc@foo arg1 arg2 arg3

为了能够向函数传递对象作为实参，引入了workspace的概念：workspace是一个路径，下边存放pickle格式的串行化对象。对于传递给函数的实参arg_n，script脚本首先检查workspace下是否存在同名文件，若存在就从该文件读取对象，若不存在，script脚本尝试把arg_n作为Python表达式来求值，并把该值作为函数实参，若求值失败，则arg_n作为字符串处理。可以通过等号操作向workspace内的变量赋值（实际上是把值保存到workspace下的文件）

./script foofunc@foo argv1 argv2 argv3 = retval

小工具由两个短文件实现：

script脚本，python函数的命令行入口

#!/usr/bin/env python
import sys
import cPickle as pickle

from utils import workspace as ws

sargv = sys.argv[1:]
argv = []
outfile = ''

try:
	eqind = sargv.index('=')
	argv = sargv[:eqind]
	outfile = sargv[eqind+1]
except:
	argv = sargv
	outfile = ''

re = ws.command(argv)

if not outfile:
    sys.stdout.write(pickle.dumps(re))
else:
    ws.save(re, outfile)

workspace支持workspace.py

import cPickle as pickle
import os

from utils import tlog

LOG = tlog.get(__name__)

def save(value, name):
    with open(os.path.join('workspace',name), 'w') as f:
        pickle.dump(value, f)
    LOG.info("Saving workspace variable: %s", name)

def load(name):
    with open(os.path.join('workspace',name)) as f:
        value = pickle.load(f)
    LOG.info("Loading workspace variable: %s",name)
    return value

def wseval(x):
    if os.path.isfile(os.path.join('workspace', x)):
        return load(x)
    try:
        re = eval(x)
        LOG.info("Eval code: %s",x)
        return re
    except:
        LOG.info("String argument: %s",x)
        return x

def command(argv):
    funname, modname = argv[0].split('@')
    mod = __import__(modname, fromlist=[argv[0]])
    fun = eval('mod.'+funname)
    try:
        nkw = filter(lambda x:not x.startswith('--'), argv[1:])
        nkw = map(wseval, nkw)
    except:
        nkw = []
    try:
        kws = filter(lambda x: x.startswith('--'), argv[1:])
        kws = map(lambda x: x.replace('--', '').split('='), kws)
        kws = dict( map(lambda x:tuple(x[0],wsevalx[1]), kws) )
    except:
        kws = {}
    if callable(fun):
        LOG.info("Excuting %s",argv)
        re = apply(fun, nkw, kws)
        LOG.info("Finish excute %s", argv)
        return re
    else:
        return fun