数据集生成的要素数据缺少必需的输入键'dense_Dense1_input'

时间:2020-08-04 21:44:57

标签: tensorflow.js

我正在尝试从CSV中加载任何适合模型的数据。这是(简化的)代码:

import { data, layers, sequential } from '@tensorflow/tfjs-node';

const dataset = data.csv('file://C:\\Users\\rgott\\Projects\\tensorflow-js-node\\data.csv', {
    columnConfigs: {
        PurchasePrice: {
            isLabel: true,
            dtype: 'float32',
        },
        LivingArea: {
            dtype: 'float32',
            required: true,
        },
        GardenArea: {
            dtype: 'float32',
            default: 0,
        },
    },
    configuredColumnsOnly: true,
    hasHeader: true,
    delimiter: ',',
});

const model = sequential({
    layers: [
        layers.dense({inputShape: [2], units: 5, activation: 'relu'}),
        layers.dense({units: 1, activation: 'softmax'}),
    ],
});
model.compile({optimizer: 'adam', loss: 'meanSquaredError'})
await model.fitDataset(dataset, {epochs: 10});

数据如下:

PurchasePrice,LivingArea,GardenArea,HasBalcony,HasElevator,HasTerrace,IncludedParkingSpots,Rooms,YearOfConstruction,FirstTimeUse
"178,695","35.5",,Yes,Yes,Yes,0,2,,Yes
"315,000",112,,No,No,No,0,,,No
"589,000","124.33","54.69",No,No,Yes,0,5,,Yes
"387,030","79.01",,Yes,Yes,Yes,0,3,,Yes
...

我遇到标题中提到的错误:

UnhandledPromiseRejectionWarning:错误:数据集生成的要素数据缺少必需的输入键'dense_Dense1_input'。

我已经搜索了Google和SO几个小时,但我不知道自己在做什么错。为什么要寻找某种输入名称?难道不就将Xs中的所有值都作为输入吗?

以防万一:

  • @tensorflow/tfjs-node2.1.0
  • ts-node8.10.2
  • typescript3.9.7
  • Windows 10
  • 节点10.16

0 个答案:

没有答案