余弦退火学习率衰减策略

最新推荐文章于 2024-10-07 06:30:00 发布

深度学习的奋斗者

最新推荐文章于 2024-10-07 06:30:00 发布

阅读量1.6k

点赞数

分类专栏：轴承故障诊断

本文链接：https://blog.csdn.net/qq_40840797/article/details/119537808

版权

轴承故障诊断专栏收录该内容

11 篇文章 141 订阅 ¥59.90 ¥99.00

订阅专栏

超级会员免费看

import tensorflow as tf
import math
import matplotlib.pyplot as plt

class CosineWarmup(tf.keras.optimizers.schedules.LearningRateSchedule):
    def __init__(self, warmup_slope, warmup_steps, cosine_steps):
        super().__init__()
        self.warmup_slope = tf.cast(warmup_slope, dtype=tf.float32)
        self.warmup_steps = tf.cast(warmup_steps, dtype=tf.float32)
        self.cosine_steps = tf.cast(cosine_steps, dtype=tf.float32)
        self.steps = self.warmup_steps + self.cosine_steps

    def __call__(self, step):
        if step%self.steps < self.warmup_steps:
            return self.warmup_slope * (step%s

了解本专栏