Inheritance diagram for fastreid.solver.optim.sgd.SGD:
Public Member Functions | |
| __init__ (self, params, lr=required, momentum=0, dampening=0, weight_decay=0, nesterov=False) | |
| __setstate__ (self, state) | |
| step (self, closure=None) | |
Detailed Description
Implements stochastic gradient descent (optionally with momentum).
Nesterov momentum is based on the formula from
`On the importance of initialization and momentum in deep learning`__.
Args:
params (iterable): iterable of parameters to optimize or dicts defining
parameter groups
lr (float): learning rate
momentum (float, optional): momentum factor (default: 0)
weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
dampening (float, optional): dampening for momentum (default: 0)
nesterov (bool, optional): enables Nesterov momentum (default: False)
Example:
>>> optimizer = torch.optim.SGD(model.parameters(), lr=0.1, momentum=0.9)
>>> optimizer.zero_grad()
>>> loss_fn(model(input), target).backward()
>>> optimizer.step()
__ http://www.cs.toronto.edu/%7Ehinton/absps/momentum.pdf
.. note::
The implementation of SGD with Momentum/Nesterov subtly differs from
Sutskever et. al. and implementations in some other frameworks.
Considering the specific case of Momentum, the update can be written as
.. math::
\begin{aligned}
v_{t+1} & = \mu * v_{t} + g_{t+1}, \\
p_{t+1} & = p_{t} - \text{lr} * v_{t+1},
\end{aligned}
where :math:`p`, :math:`g`, :math:`v` and :math:`\mu` denote the
parameters, gradient, velocity, and momentum respectively.
This is in contrast to Sutskever et. al. and
other frameworks which employ an update of the form
.. math::
\begin{aligned}
v_{t+1} & = \mu * v_{t} + \text{lr} * g_{t+1}, \\
p_{t+1} & = p_{t} - v_{t+1}.
\end{aligned}
The Nesterov version is analogously modified.
Constructor & Destructor Documentation
◆ __init__()
| fastreid.solver.optim.sgd.SGD.__init__ | ( | self, | |
| params, | |||
| lr = required, | |||
| momentum = 0, | |||
| dampening = 0, | |||
| weight_decay = 0, | |||
| nesterov = False ) |
Definition at line 44 of file sgd.py.
45 weight_decay=0, nesterov=False):
46 if lr is not required and lr < 0.0:
47 raise ValueError("Invalid learning rate: {}".format(lr))
48 if momentum < 0.0:
49 raise ValueError("Invalid momentum value: {}".format(momentum))
50 if weight_decay < 0.0:
51 raise ValueError("Invalid weight_decay value: {}".format(weight_decay))
52
53 defaults = dict(lr=lr, momentum=momentum, dampening=dampening,
54 weight_decay=weight_decay, nesterov=nesterov)
55 if nesterov and (momentum <= 0 or dampening != 0):
56 raise ValueError("Nesterov momentum requires a momentum and zero dampening")
57 super(SGD, self).__init__(params, defaults)
58
Member Function Documentation
◆ __setstate__()
| fastreid.solver.optim.sgd.SGD.__setstate__ | ( | self, | |
| state ) |
◆ step()
| fastreid.solver.optim.sgd.SGD.step | ( | self, | |
| closure = None ) |
Performs a single optimization step.
Arguments:
closure (callable, optional): A closure that reevaluates the model
and returns the loss.
Definition at line 65 of file sgd.py.
65 def step(self, closure=None):
66 """Performs a single optimization step.
67 Arguments:
68 closure (callable, optional): A closure that reevaluates the model
69 and returns the loss.
70 """
71 loss = None
72 if closure is not None:
73 with torch.enable_grad():
74 loss = closure()
75
76 for group in self.param_groups:
77 if group['freeze']: continue
78
79 weight_decay = group['weight_decay']
80 momentum = group['momentum']
81 dampening = group['dampening']
82 nesterov = group['nesterov']
83
84 for p in group['params']:
85 if p.grad is None:
86 continue
87 d_p = p.grad
88 if weight_decay != 0:
89 d_p = d_p.add(p, alpha=weight_decay)
90 if momentum != 0:
91 param_state = self.state[p]
92 if 'momentum_buffer' not in param_state:
93 buf = param_state['momentum_buffer'] = torch.clone(d_p).detach()
94 else:
95 buf = param_state['momentum_buffer']
96 buf.mul_(momentum).add_(d_p, alpha=1 - dampening)
97 if nesterov:
98 d_p = d_p.add(buf, alpha=momentum)
99 else:
100 d_p = buf
101
102 p.add_(d_p, alpha=-group['lr'])
103
104 return loss
The documentation for this class was generated from the following file:
- smreid/fastreid/solver/optim/sgd.py
