NegativeLogLikelihood

1 /++ equivalent torch v0.4 code
2  >>> x = torch.FloatTensor([[0.2, 0.4, 0.4], [0.1,0.5,0.4]])
3  >>> x.requires_grad = True
4  >>> t = torch.LongTensor([1, 0])
5  >>> l = torch.nn.functional.nll_loss(x, t)
6  >>> print(l)
7  tensor(-0.2500)
8 
9  >>> l.backward()
10  >>> print(x.grad)
11  tensor([[0.0, -0.5, 0.0], [-0.5, 0.0, 0.0]])
12  +/
13 import std.typecons;
14 import grain.testing;
15 
16 NegativeLogLikelihood!(float, int) func;
17 auto hx = [[0.2f, 0.4f, 0.4f], [0.1f, 0.5f, 0.4f], [0.1f, 0.5f, 0.4f]]
18     .variable;
19 auto ht = [1, 0, func.ignoreIndex].variable;
20 auto hl = func.forward(hx, ht);
21 assert(func._normalize == 0.5);
22 assert(hl.sliced == [-(0.4f + 0.1f + 0.0f) / 2]);
23 auto hgx = func.backward(1.0f.variable);
24 assert(hgx[0].sliced == [[0.0, -0.5, 0.0], [-0.5, 0.0, 0.0], [0.0, 0.0, 0.0]]);
25 assert(!hgx[1].defined);
26 gradCheck(func, tuple(hx, ht), 1.0f.variable);
27 
28 version (grain_cuda) {
29     auto dx = hx.to!DeviceStorage;
30     auto dt = ht.to!DeviceStorage;
31     auto dl = func.forward(dx, dt);
32     assert(func._normalize == 0.5);
33     assert(dl.to!HostStorage.sliced == [-(0.4f + 0.1f + 0.0f) / 2]);
34     auto dgx = func.backward(1.0f.variable.to!DeviceStorage);
35     assert(dgx[0].to!HostStorage.sliced ==
36            [[0.0, -0.5, 0.0],
37             [-0.5, 0.0, 0.0],
38             [0.0, 0.0, 0.0]]);
39     assert(!dgx[1].defined);
40 }

1 import std.typecons;
2 import grain.testing;
3 import mir.ndslice;
4 static import grain.config;
5 
6 grain.config.backprop = true;
7 
8 NegativeLogLikelihood!(float, int) func;
9 auto hx = [[0.2f, 0.4f, 0.4f], [0.1f, 0.5f, 0.4f], [0.1f, 0.5f, 0.4f]]
10     .variable;
11 hx.requiresGrad = true;
12 auto ht = [1, 0, func.ignoreIndex].variable;
13 auto hl = func.applyForward(hx, ht);
14 
15 assert(func._normalize == 0.5);
16 assert(hl.sliced == [-(0.4f + 0.1f + 0.0f) / 2]);
17 auto u = UntypedVariable(1.0f.variable);
18 hl.backward(&u);
19 
20 assert(hx.grad[].sliced(3, 3) == [[0.0, -0.5, 0.0], [-0.5, 0.0, 0.0], [0.0, 0.0,
21         0.0]]);
22 // TODO assert(!ht.grad.defined);