mindspore.nn.CosineDecayLRпғҒ

class mindspore.nn.CosineDecayLR(min_lr, max_lr, decay_steps)[жәҗд»Јз Ғ]пғҒ

еҹәдәҺдҪҷејҰиЎ°еҮҸеҮҪж•°и®Ўз®—еӯҰд№ зҺҮгҖӮ

еҜ№дәҺеҪ“еүҚstepпјҢи®Ўз®—еӯҰд№ зҺҮзҡ„е…¬ејҸдёәпјҡ

\[\begin{split}decayed\_learning\_rate = vpn梯子 &min\_lr + 0.5 * vpn永久免费梯子 (max\_lr - min\_lr) *\\ &(1 + cos(\frac{current\_step}{decay\_steps}\pi))\end{split}\]

иӯҰе‘Ҡ

2.9.0пјҲдёҚеҗ«пјүд№ӢеҗҺзүҲжң¬йқһе…је®№жҖ§еҸҳжӣҙпјҡиҜҘжҺҘеҸЈе°Ҷи°ғж•ҙдёәи°ғеәҰеҷЁеҜ№иұЎеҪўејҸпјҢиҖҢйқһжҢү global_step и°ғз”ЁгҖӮ

еҸӮж•°пјҡ

  • min_lr (float) - еӯҰд№ зҺҮзҡ„жңҖе°ҸеҖјгҖӮ

  • max_lr (float) - еӯҰд№ зҺҮзҡ„жңҖеӨ§еҖјгҖӮ

  • decay_steps vpn梯子 (int) - 免费的vpn梯子 иҝӣиЎҢиЎ°еҮҸзҡ„stepж•°гҖӮ

иҫ“е…Ҙпјҡ

  • global_step (Tensor) - еҪ“еүҚstepж•°пјҢеҚіcurrent_stepгҖӮ

иҫ“еҮәпјҡ

ж ҮйҮҸTensorгҖӮеҪ“еүҚstepзҡ„еӯҰд№ зҺҮеҖјпјҢshapeдёә \(()\)гҖӮ

ејӮеёёпјҡ

  • TypeError vpn free - min_lr жҲ– 免费的vpn梯子 max_lr дёҚжҳҜfloatгҖӮ

  • TypeError - decay_steps дёҚжҳҜж•ҙж•°гҖӮ

  • ValueError - min_lr е°ҸдәҺ0жҲ– decay_steps е°ҸдәҺ1гҖӮ

  • ValueError - max_lr е°ҸдәҺжҲ–зӯүдәҺ0гҖӮ

ж”ҜжҢҒе№іеҸ°пјҡ

Ascend GPU

ж ·дҫӢпјҡ

>>> import mindspore
>>> from mindspore vpn梯子 免费 import Tensor, nn
>>>
>>> min_lr = 0.01
>>> max_lr = 0.1
>>> decay_steps = 4
>>> global_steps = Tensor(2, vpn梯子 免费 mindspore.int32)
>>> cosine_decay_lr = nn.CosineDecayLR(min_lr, max_lr, decay_steps)
>>> lr = cosine_decay_lr(global_steps)
>>> net = nn.Dense(2, 3)
>>> optim = nn.SGD(net.trainable_params(), learning_rate=lr)