convnetjs/demo/getting-started/regression.html at 5fa20a55acebe49f0d26311c4d0d81cea603e3f1 · karpathy/convnetjs · 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
<!doctype html>
<html>
  <head>
    <title>
      Hello wurl
    </title>
    <body>
      <script src="../../build/convnet-min.js"></script>
      <script>
        //for regression as opposed to classification
        //we can proceed similarly just replace the classifier
        //layer from softmax
        const layer_defs = []
        layer_defs.push({
          type: 'input',
          out_sx: 1,
          out_sy: 1,
          out_depth: 2
        })
        layer_defs.push({
          type: 'fc',
          num_neurons: 5,
          activation: 'sigmoid'
        })
        layer_defs.push({
          type: 'regression',
          num_neurons: 1
        })

        const net = new convnetjs.Net()
        net.makeLayers(layer_defs)
        const x = new convnetjs.Vol([0.5, -1.3])
        const trainer = new convnetjs.SGDTrainer(net, {
          learning_rate: 0.01,
          momentum: 0.0,
          batch_size: 1,
          l2_decay: 0.001
        })
        //train on the datapoint, saying [0.5, -1.3]
        //should map to value 0.7
        for (var i = 0; i < 200; i++) {
          trainer.train(Object.assign({}, x), [0.7])
        }

        const predicted_values = net.forward(x)
        console.log(predicted_values.w)
      </script>
    </body>
  </head>
</html>