{"id":206,"date":"2014-07-09T23:54:26","date_gmt":"2014-07-09T14:54:26","guid":{"rendered":"https:\/\/www.dogrow.net\/nnet\/?p=206"},"modified":"2025-06-11T22:50:58","modified_gmt":"2025-06-11T13:50:58","slug":"blog16","status":"publish","type":"post","link":"https:\/\/www.dogrow.net\/nnet\/blog16\/","title":{"rendered":"(16) cuda-convnet\u306e\u7d50\u679c\u304b\u3089error matrix\u3092\u4f5c\u6210"},"content":{"rendered":"<h1 class=\"my_h\">1. \u5168\u30c6\u30b9\u30c8\u7d50\u679c\u3092\u53d6\u5f97\u3059\u308b\u3002<\/h1>\n<p><span class=\"my_fc_deeppinkB\">cuda-convnet <\/span>\u306e\u30c6\u30b9\u30c8\u7d50\u679c\u51fa\u529b\u6a5f\u80fd\u3092\u4f7f\u3046\u3068\u3001\u5404\u30c6\u30b9\u30c8\u753b\u50cf\u306b\u3064\u3044\u3066\u51fa\u529b\u5c64\u306e\u5168\u30e6\u30cb\u30c3\u30c8\u306e\u51fa\u529b\u5024\u304c\u53d6\u5f97\u3067\u304d\u308b\u3002<br \/>\n\u4f8b\u3048\u3070\u3001\u5b66\u7fd2\u7d50\u679c <span class=\"my_fc_deeppinkB\">ConvNet__2014-07-09_22.31.05<\/span> \u304c\u5b58\u5728\u3057\u3066\u3044\u308b\u5834\u5408\u3001\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u53d6\u5f97\u3067\u304d\u308b\u3002<\/p>\n<pre>[user@linux]$ python shownet.py -f .\/save\/ConvNet__2014-07-09_22.31.05 --write-features=probs --feature-path=..\/tmp\/\r\n<\/pre>\n<p>\u5b9f\u884c\u3057\u305f\u7d50\u679c\u3001<span class=\"my_fc_blueB\">&#8211;feature-path <\/span>\u3067\u6307\u5b9a\u3057\u305f\u30c7\u30a3\u30ec\u30af\u30c8\u30ea\u306b\u30c6\u30b9\u30c8\u7d50\u679c\u304c\u51fa\u529b\u3055\u308c\u308b\u3002<\/p>\n<pre>[user@linux]$ ll\r\n\u5408\u8a08 436\r\n-rw-rw-r--. 1 user user     78  7\u6708  9 22:54 2014 batches.meta\r\n-rw-rw-r--. 1 user user 440191  7\u6708  9 22:54 2014 data_batch_6\r\n<\/pre>\n<p>\u307e\u305a\u306f\u51fa\u529b\u3055\u308c\u305f <span class=\"my_fc_deeppinkB\">batches.meta<\/span> \u306e\u4e2d\u8eab\u3092\u898b\u3066\u307f\u308b\u3002<\/p>\n<pre>&gt;&gt;&gt; import cPickle as cp\r\n&gt;&gt;&gt; mt = cp.load(open('batches.meta'))\r\n&gt;&gt;&gt; mt.keys()\r\n['num_vis', 'source_model']\r\n&gt;&gt;&gt; mt['num_vis']\r\n10\r\n<\/pre>\n<p>num_vis \u304b\u3089\u30e9\u30d9\u30eb\u6570\u304c\u53d6\u5f97\u3067\u304d\u308b\u3088\u3046\u3060\u3002<\/p>\n<p>\u6b21\u306b <span class=\"my_fc_deeppinkB\">data_batch_6<\/span> \u306e\u4e2d\u8eab\u3092\u898b\u3066\u307f\u308b\u3002<\/p>\n<pre>&gt;&gt;&gt; sv = cp.load(open('data_batch_6'))\r\n&gt;&gt;&gt; sv.keys()\r\n['labels', 'data']\r\n&gt;&gt;&gt; sv['labels']\r\narray([[ 3.,  8.,  8., ...,  5.,  1.,  7.]], dtype=float32)\r\n&gt;&gt;&gt; sv['data']\r\narray([[  2.12459755e-03,   2.84853554e-03,   8.47100373e-03, ...,\r\n          1.67425780e-03,   6.06818078e-03,   1.81574584e-03],\r\n       [  9.24929883e-03,   9.61963892e-01,   7.97947752e-04, ...,\r\n          7.32670560e-06,   2.74835732e-02,   3.86457832e-04],\r\n       ...,\r\n       [  6.13748282e-02,   3.04950565e-01,   1.83501482e-01, ...,\r\n          1.76601484e-02,   1.54394070e-02,   5.21140127e-03],\r\n       [  7.03860400e-03,   1.66299902e-02,   1.75639763e-02, ...,\r\n          6.83449626e-01,   1.96132925e-03,   3.77259240e-03]], dtype=float32)\r\n<\/pre>\n<p>labels \u306b\u306f\u6b63\u89e3\u30e9\u30d9\u30eb\u3001data \u306b\u306f\u5404\u753b\u50cf\u306e\u51fa\u529b\u5c64\u5168\u30e6\u30cb\u30c3\u30c8\u51fa\u529b\u5024\u304c\u66f8\u304b\u308c\u3066\u3044\u308b\u3002<\/p>\n<h1 class=\"my_h\">2. error matrix\u3092\u4f5c\u6210\u3059\u308b\u3002<\/h1>\n<p>\u4e0a\u8a18(1)\u3067\u4f5c\u6210\u3057\u305f\u30c7\u30fc\u30bf\u304b\u3089 <span class=\"my_fc_deeppinkB\">error matrix<\/span> \u3092\u4f5c\u6210\u3059\u308b\u95a2\u6570\u306f\u4ee5\u4e0b\u306e\u901a\u308a\u3002<br \/>\n<span class=\"my_fc_gray\">\u62d9\u3044Python\u30b9\u30ad\u30eb\u3067\u8a66\u884c\u932f\u8aa4\u3057\u306a\u304c\u3089\u66f8\u3044\u305f\u306e\u3067\u7121\u99c4\u304c\u3044\u3063\u3071\u3044\u3042\u308b\u304b\u3082\u2026<\/span><\/p>\n<pre>import cPickle as cp\r\nimport numpy as np\r\n\r\ndef makeErrorMatrix( metaFile, dataFile ):\r\n    mt = cp.load(open(metaFile))\r\n    sv = cp.load(open(dataFile))\r\n    nLabel = mt['num_vis']\r\n    nImage = sv['labels'].shape[1]\r\n\r\n    print '# of label : %d' % nLabel\r\n    print '# of image : %d' % nImage\r\n\r\n    maxAryRaw = np.uint8(np.argmax(sv['data'], 1))\r\n    maxArySpv = np.uint8(sv['labels'].T.flatten())\r\n\r\n    errorMatrix = np.zeros((nLabel,nLabel),dtype=int)\r\n    for i in range(0,nImage):\r\n        x = maxAryRaw[i]\r\n        y = maxArySpv[i]\r\n        errorMatrix[y][x] = errorMatrix[y][x] + 1\r\n    np.savetxt('errmtx.csv', errorMatrix, fmt='%d', delimiter=',')\r\n<\/pre>\n<p><span class=\"my_fc_deeppinkB\">CIFAR-10 <\/span>\u306e\u30c6\u30b9\u30c8\u7d50\u679c\u304b\u3089\u4f5c\u6210\u3057\u305fCSV\u30d5\u30a1\u30a4\u30eb\u306f\u3053\u308c\uff08\u2193\uff09<\/p>\n<pre>738, 42,120, 13, 13,  6,  5, 10, 28, 25\r\n 28,842, 21, 14, 11,  7,  2,  2,  5, 68\r\n 68, 20,599, 56, 95, 99, 33, 16,  6,  8\r\n 39, 31,114,466, 66,193, 48, 24,  7, 12\r\n 44, 12,126, 55,572, 83, 32, 64,  4,  8\r\n 25, 11, 87,214, 38,571, 11, 30,  4,  9\r\n  7, 15,103, 85, 92, 53,633,  3,  5,  4\r\n 18,  8, 67, 63, 72,102,  3,644,  0, 23\r\n145, 91, 62, 27,  7,  9,  3,  3,627, 26\r\n 54,190, 11, 12,  7, 11,  5, 19, 14,677\r\n<\/pre>\n<p>MS-Excel\u3067\u6574\u5f62\u3059\u308b\u3068\u3001\u305d\u308c\u3063\u307d\u3044 <span class=\"my_fc_deeppinkB\">error matrix<\/span> \u304c\u51fa\u6765\u4e0a\u304c\u308b\u3002<br \/>\n\u4eca\u56de\u306e\u5206\u985e\u7d50\u679c\u3092\u898b\u308b\u3068&#8230;<br \/>\n1) recall\u3092\u898b\u308b\u3068automobile\u306e\u6b63\u89e3\u7387\u304c\u7a81\u51fa\u3057\u3066\u9ad8\u3044\u3088\u3046\u306b\u898b\u3048\u308b\u304c\u3001precision\u3092\u898b\u308b\u3068\u6b63\u78ba\u3055\u306f\u5e73\u5747\u4e26\u307f\u3002<br \/>\n\u3000\u3000\u2192 \u5168\u30c6\u30b9\u30c8\u30c7\u30fc\u30bf\u4e2d\u3067automobile\u3068\u8a8d\u8b58\u3055\u308c\u308b\u6570\u306f\u591a\u3044\u304c\u3001\u5b9f\u969b\u306bautomobile\u3067\u3042\u308b\u78ba\u7387\u306f\u4e26<br \/>\n2) recall\u3092\u898b\u308b\u3068ship\u306e\u6b63\u89e3\u7387\u306f\u5e73\u5747\u4e26\u307f\u3060\u304c\u3001precision\u3092\u898b\u308b\u3068\u6b63\u78ba\u3055\u306f\u7a81\u51fa\u3057\u3066\u3044\u308b\u3002<br \/>\n\u3000\u3000\u2192 ship\u3068\u8a8d\u8b58\u3055\u308c\u305f\u3068\u304d\u306b\u3001\u672c\u5f53\u306bship\u3067\u3042\u308b\u78ba\u7387\u304c\u9ad8\u3044\u3002<br \/>\n3) truck\u3092automobile\u3068\u3088\u304f\u9593\u9055\u3048\u308b\u3002<br \/>\n\u3000\u3000\u2192 \u5199\u771f\u3060\u3051\u898b\u305f\u3089\u78ba\u304b\u306b\u4f3c\u3066\u308b\u304b&#8230;<br \/>\n4) cat\u3068dog\u306f\u304a\u4e92\u3044\u306b\u3088\u304f\u9593\u9055\u3048\u308b\u3002<br \/>\n\u3000\u3000\u2192 \u3053\u308c\u3082\u30b7\u30eb\u30a8\u30c3\u30c8\u3060\u3051\u898b\u305f\u3089\u533a\u5225\u304c\u3064\u304b\u306a\u3044\u304b\u3082&#8230;<br \/>\n\u306a\u3069\u306e\u7279\u5fb4\u304c\u898b\u3089\u308c\u3066\u9762\u767d\u3044\u3002<br \/>\n<a href=\"https:\/\/www.dogrow.net\/nnet\/wp-content\/uploads\/2014\/07\/20140709_05.png\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/www.dogrow.net\/nnet\/wp-content\/uploads\/2014\/07\/20140709_05.png\" alt=\"20140709_05\" width=\"797\" height=\"308\" class=\"alignnone size-full wp-image-221\" srcset=\"https:\/\/www.dogrow.net\/nnet\/wp-content\/uploads\/2014\/07\/20140709_05.png 797w, https:\/\/www.dogrow.net\/nnet\/wp-content\/uploads\/2014\/07\/20140709_05-300x115.png 300w\" sizes=\"auto, (max-width: 797px) 100vw, 797px\" \/><\/a><\/p>\n<hr class=\"my_hr_bottom\">\n","protected":false},"excerpt":{"rendered":"<p>1. \u5168\u30c6\u30b9\u30c8\u7d50\u679c\u3092\u53d6\u5f97\u3059\u308b\u3002 cuda-convnet \u306e\u30c6\u30b9\u30c8\u7d50\u679c\u51fa\u529b\u6a5f\u80fd\u3092\u4f7f\u3046\u3068\u3001\u5404\u30c6\u30b9\u30c8\u753b\u50cf\u306b\u3064\u3044\u3066\u51fa\u529b\u5c64\u306e\u5168\u30e6\u30cb\u30c3\u30c8\u306e\u51fa\u529b\u5024\u304c\u53d6\u5f97\u3067\u304d\u308b\u3002 \u4f8b\u3048\u3070\u3001\u5b66\u7fd2\u7d50\u679c ConvNet__2014-07-09_22.31.\u2026 <span class=\"read-more\"><a href=\"https:\/\/www.dogrow.net\/nnet\/blog16\/\">\u7d9a\u304d\u3092\u8aad\u3080 &raquo;<\/a><\/span><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8,10],"tags":[],"class_list":["post-206","post","type-post","status-publish","format-standard","hentry","category-cuda","category-cuda-convnet"],"views":2493,"amp_enabled":true,"_links":{"self":[{"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/posts\/206","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/comments?post=206"}],"version-history":[{"count":21,"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/posts\/206\/revisions"}],"predecessor-version":[{"id":2548,"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/posts\/206\/revisions\/2548"}],"wp:attachment":[{"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/media?parent=206"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/categories?post=206"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.dogrow.net\/nnet\/wp-json\/wp\/v2\/tags?post=206"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}