blob: 9dc53860e526133e9117bd33ae603c2526b60c25 (
plain)
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
51
52
53
54
55
56
57
58
|
op {
graph_op_name: "SparseApplyProximalGradientDescent"
in_arg {
name: "var"
description: <<END
Should be from a Variable().
END
}
in_arg {
name: "alpha"
description: <<END
Scaling factor. Must be a scalar.
END
}
in_arg {
name: "l1"
description: <<END
L1 regularization. Must be a scalar.
END
}
in_arg {
name: "l2"
description: <<END
L2 regularization. Must be a scalar.
END
}
in_arg {
name: "grad"
description: <<END
The gradient.
END
}
in_arg {
name: "indices"
description: <<END
A vector of indices into the first dimension of var and accum.
END
}
out_arg {
name: "out"
description: <<END
Same as "var".
END
}
attr {
name: "use_locking"
description: <<END
If True, the subtraction will be protected by a lock;
otherwise the behavior is undefined, but may exhibit less contention.
END
}
summary: "Sparse update \'*var\' as FOBOS algorithm with fixed learning rate."
description: <<END
That is for rows we have grad for, we update var as follows:
$$prox_v = var - alpha * grad$$
$$var = sign(prox_v)/(1+alpha*l2) * max{|prox_v|-alpha*l1,0}$$
END
}
|