cnnsv/data.lua at master · simpleoier/cnnsv · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
math.randomseed(os.time())
----------------------------------------------------------------------
-- function readfbank(filename, lines)
--    local file = io.open(filename, 'r')
--    if not(file) then
--       print("Error "..filename.." does not exist")
--       return 0
--    end
--    for line in file:lines() do
--       lines[#lines+1] = line
--    end
--    file:close()
--    return lines
-- end

-- function parsefbank(lines, data, labels)
--    for i = 1,#lines do
--       tb = {}
--       for k, v in string.gmatch(lines[i], "%S*") do
--          if 0~=string.len(k) then
--             tb[#tb+1] = tonumber(k)
--          end
--       end
--       labels[i] = tb[1]

--       -- r = math.random()
--       -- if (r<0.3) then
--       --    labels[i] = 2
--       --    sum = sum+1
--       -- end

--       for h = 1, height do
--          for m = 1, nfeats do
--             for w = 1, width do
--                data[i][m][h][w] = tb[1+w+(m-1)*width+(h-1)*nfeats*width]
--             end
--          end
--       end
--    end
-- end

function readLabel(filename)
   filelabel = {}
   nlabels = 0
   if (filename~='') then
      local fin = io.open(filename,'r')
      for line in fin:lines() do
         local l = line:split(' ')
         filelabel[l[1]] = tonumber(l[2])
         nlabels = nlabels+1
      end
   end
end

-- function readDataFeat(featfile)
--    print('==> reading feature from '..opt.featfile)
--    local lines = {}
--    local curlinenum = 0
--    while (curlinenum<opt.maxrows) do
--       curlinenum = curlinenum+1
--       local line = featfile:read()
--       if (line~=nil) then
--          lines[#lines+1] = line
--       else
--          break
--       end
--    end

--    local tdata = torch.Tensor(#lines, ninputs)
--    local tlabels = torch.Tensor(#lines,1)

--    parsefbank(lines, tdata, tlabels)
-- -- print(trlabels)
--    local Data = {
--       data = tdata,
--       labels = tlabels,
--       size = function() return #lines end
--    }
--    return Data
-- end

-- function readDataScp1(listfile,filenum)
--    local lines = {}
--    local curlinenum = 0
--    while (curlinenum<filenum) do
--       curlinenum = curlinenum+1
--       local line = listfile:read()
--       if (line~=nil) then
--          local fbankfilename = line
--          print("Reading feature from "..fbankfilename)
--          readfbank(fbankfilename, lines)
--       else
--          break
--       end
--    end

--    local tdata = torch.Tensor(#lines, nfeats, height, width)
--    local tlabels = torch.Tensor(#lines,1)

--    parsefbank(lines, tdata, tlabels)
-- -- print(trlabels)
--    local Data = {
--       data = tdata,
--       labels = tlabels,
--       size = function() return #lines end
--    }
--    return Data
-- end

function readDataScp2(listfile,filenum,means,variances)
   local feats = {}
   local labels = {}
   local curlinenum = 0
   while (curlinenum<filenum) do
      curlinenum = curlinenum+1
      local line = listfile:read()
      if (line~=nil) then
         filename = paths.basename(line)
         local chunk = filename:split('_')
         print("Reading feature from "..line)
         assert(io.open(line),"File " .. line .. " does not exist!")
         local feat = loadhtk(line, frameExt)
         for i=1,feat:size(1) do
            feats[#feats+1] = feat[i]
            if (nlabels~=0) then
               labels[#feats] = filelabel[chunk[1]]
            else
               labels[#feats] = 1
            end
         end
      else
         break
      end
   end

   if (#feats==0) then
      return
   end
   -- print(labels)
   local tdata = torch.Tensor(#feats, feats[1]:size(1))
   local tlabels = torch.Tensor(labels)
   for i=1,#feats do
      tdata[i] = feats[i]
      -- normlize if globalnorm exists, T norm is defined as (data-means)/variance
      if (opt.globalnorm~='') then
         tdata[i]:map2(means, variances, function(data,mean,variance) return (data+mean)*variance end)
      end
   end

   local Data = {
      data = tdata,
      labels = tlabels,
      size = function() return #feats end
   }
   return Data
end