Question

我训练了一个中立的网络，以根据页边距中间的距离和角度预测页面上某个位置的信息。但是对于我没有训练过的数据进行预测，当使用该数据运行网络时，训练模型中没有任何距离和角度，我猜只是随机输出？

下面是一个例子：

{ line: 'Tax', predict: [ sub_total: 0.9999998807907104 ] }
{ line: '$9.50', predict: [ tax: 0.9999759197235107 ] }
{ line:
   'Payment is due within 30 days from date of invoice. Late payment is subject to fees of 5% per month.',
  predict: [ sub_total: 0.9999998807907104 ] }
{ line: '', predict: [ sub_total: 0.9999998807907104 ] }
{ line: '', predict: [ sub_total: 0.9999998807907104 ] }
{ line: '', predict: [ sub_total: 0.9999998807907104 ] }
{ line: 'info@slicedinvoices.com',
  predict: [ sub_total: 0.9999998807907104 ] }
{ line: 'slicedinvoices.com',
  predict: [ sub_total: 0.9999998807907104 ] }
{ line: '555 1234 123',
  predict: [ sub_total: 0.9999998807907104 ] }
{ line: 'Page 1/1', predict: [ sub_total: 0.9998056888580322 ] }

所以下面您可以看到，所有这些数据都没有在训练模型中，我没有训练过以进行预测，$ 9.5的税款是完全正确的预测，但无论在哪里，其他数据都将预测为小计它放在屏幕上... 无论如何要修复brain.js错误地预测数据？我的训练模型具有以下结构：

{
        "key": "title",
        "value": "Invoice",
        "top": [
            187.48657949037312,
            167.1073362689553
        ],
        "bottom": [
            311.3688900645021,
            -125.94132973440453
        ],
        "left": [
            491.7684023117488,
            -167.65624617099695
        ],
        "right": [
            155.72223961993998,
            -42.46212375329803
        ]
    }

然后我将这些值除以1000，然后将它们设为正数，从而使这些值实际上相差很远...

Brain.js错误的预测=

0 个答案: