Torch 日志文件的保存 logroll

最新推荐文章于 2024-03-10 16:08:41 发布

a1424262219

最新推荐文章于 2024-03-10 16:08:41 发布

阅读量323

点赞数

原文链接：http://www.cnblogs.com/wangxiaocvpr/p/6131602.html

版权

Torch 日志文件的保存 logroll

　　怎样将 Torch 在终端显示的信息，保存到 log 文件中？

　　现在介绍一种方法：利用 logroll 的方式。

　　参考 https://github.com/rosejn/logroll

　　 1. 首先安装 logroll 工具包：

　　　　 luarocks install logroll

　　 2. 在代码头部：require "logroll" 加载该工具包:

　　　　调用的代码如下：

 1 require 'paths'
 2 require 'os'
 3 require 'io'
 4 require 'string'
 5 
 6 require 'fn'
 7 require 'pprint'
 8 
 9 logroll = {}
10 
11 local DEFAULT_LEVEL = 'INFO'
12 local LOG_LEVELS = {'DEBUG', 'INFO', 'WARN', 'ERROR'}
13 
14 for i, label in ipairs(LOG_LEVELS) do
15     logroll[label] = i
16 end
17 logroll.levels = LOG_LEVELS
18 
19 
20 local function default_formatter(level, ...)
21     local msg = nil
22 
23     if #{...} > 1 then
24         msg = string.format(({...})[1], unpack(fn.rest({...})))
25     else
26         msg = pprint.pretty_string(({...})[1])
27     end
28 
29     return string.format("[%s - %s] - %s\n", LOG_LEVELS[level], os.date("%Y_%m_%d_%X"), msg)
30 end
31 
32 
33 local function default_writer(logger, level, ...)
34     if level >= logger.level then
35         logger.file:write(logger.formatter(level, unpack({...})))
36     end
37 end
38 
39 
40 local function make_logger(file, options)
41     local logger = {options   = options,
42                     file      = file,
43                     formatter = options.formatter or default_formatter,
44                     writer    = options.writer or default_writer,
45                     level     = logroll[DEFAULT_LEVEL]
46                 }
47 
48     return fn.reduce(function(lg, level)
49         lg[string.lower(level)] = fn.partial(logger.writer, logger, logroll[level])
50         return lg
51     end,
52     logger, LOG_LEVELS)
53 end
54 
55 
56 -- A simple logger to print to STDIO.
57 function logroll.print_logger(options)
58     local options = options or {}
59     return make_logger(io.stdout, options)
60 end
61 
62 
63 -- A logger that prints to a file.
64 function logroll.file_logger(path, options)
65     local options = options or {}
66 
67     if options.file_timestamp then
68         -- append timestamp to create unique log file
69         path = path .. '-'..os.date("%Y_%m_%d_%X")
70     end
71 
72     os.execute('mkdir -p "' .. paths.dirname(path) .. '"')
73 
74     return make_logger(io.open(path, 'w'), options)
75 end
76 
77 
78 -- A logger that combines several other loggers
79 function logroll.combine(...)
80 
81     local joint = {
82         subloggers = {...}
83     }
84 
85     for _,level in ipairs(LOG_LEVELS) do
86         local fname = string.lower(level)
87         joint[fname] = function(...)
88             for _,lg in ipairs(joint.subloggers) do
89                 lg[fname](...)
90             end
91         end
92     end
93 
94     return joint
95 end

　另外，也可以参考这个代码中，保存 log 文件的方式：

　　　https://github.com/szagoruyko/cifar.torch/blob/master/train.lua#L7

 1   if testLogger then
 2     paths.mkdir(opt.save)
 3     testLogger:add{train_acc, confusion.totalValid * 100}
 4     testLogger:style{'-','-'}
 5     testLogger:plot()
 6 
 7     if paths.filep(opt.save..'/test.log.eps') then
 8       local base64im
 9       do
10         os.execute(('convert -density 200 %s/test.log.eps %s/test.png'):format(opt.save,opt.save))
11         os.execute(('openssl base64 -in %s/test.png -out %s/test.base64'):format(opt.save,opt.save))
12         local f = io.open(opt.save..'/test.base64')
13         if f then base64im = f:read'*all' end
14       end
15 
16       local file = io.open(opt.save..'/report.html','w')
17       file:write(([[
18       <!DOCTYPE html>
19       <html>
20       <body>
21       <title>%s - %s</title>
22       <img src="https://img-blog.csdnimg.cn/2022010702433917052.png">
23       <h4>optimState:</h4>
24       <table>
25       ]]):format(opt.save,epoch,base64im))
26       for k,v in pairs(optimState) do
27         if torch.type(v) == 'number' then
28           file:write('<tr><td>'..k..'</td><td>'..v..'</td></tr>\n')
29         end
30       end
31       file:write'</table><pre>\n'
32       file:write(tostring(confusion)..'\n')
33       file:write(tostring(model)..'\n')
34       file:write'</pre></body></html>'
35       file:close()
36     end
37   end

其实，最简单有效粗暴的，应该是这种：

1     -- save the information into log files. 
2     local logSavePath = './log_Files/'
3     local file = io.open(logSavePath..'train_record.log','a')
4     file:write(tostring("train error: " .. error)..'\n')
5     file:close()

其中，第 4 行的保存记录，也可以是其他地方的变量以及语句，等等。可以随机应变。

转载于:https://www.cnblogs.com/wangxiaocvpr/p/6131602.html