SsYyH

Results 5 comments of SsYyH

> ![mygif](https://private-user-images.githubusercontent.com/123875830/239668594-51e54c97-b20f-4f74-ae53-f4f22b646d8a.gif?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MTE1NDUzODMsIm5iZiI6MTcxMTU0NTA4MywicGF0aCI6Ii8xMjM4NzU4MzAvMjM5NjY4NTk0LTUxZTU0Yzk3LWIyMGYtNGY3NC1hZTUzLWY0ZjIyYjY0NmQ4YS5naWY_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwMzI3JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDMyN1QxMzExMjNaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1hZjBjMGZjN2RiMGFiNDdlN2Y2Nzc1ZGZkY2FiZTc4Y2YwMzY5NDY5ZTY2NTY2YWU4ZmNjZTBhNmJhZmJlYTg5JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.SrsnflumboWC7_41SS_3d_d-VoPUTkDxCIB_Xyb53fA) [ ![mygif](https://private-user-images.githubusercontent.com/123875830/239668594-51e54c97-b20f-4f74-ae53-f4f22b646d8a.gif?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MTE1NDUzODMsIm5iZiI6MTcxMTU0NTA4MywicGF0aCI6Ii8xMjM4NzU4MzAvMjM5NjY4NTk0LTUxZTU0Yzk3LWIyMGYtNGY3NC1hZTUzLWY0ZjIyYjY0NmQ4YS5naWY_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwMzI3JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDMyN1QxMzExMjNaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1hZjBjMGZjN2RiMGFiNDdlN2Y2Nzc1ZGZkY2FiZTc4Y2YwMzY5NDY5ZTY2NTY2YWU4ZmNjZTBhNmJhZmJlYTg5JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.SrsnflumboWC7_41SS_3d_d-VoPUTkDxCIB_Xyb53fA) ](https://private-user-images.githubusercontent.com/123875830/239668594-51e54c97-b20f-4f74-ae53-f4f22b646d8a.gif?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MTE1NDUzODMsIm5iZiI6MTcxMTU0NTA4MywicGF0aCI6Ii8xMjM4NzU4MzAvMjM5NjY4NTk0LTUxZTU0Yzk3LWIyMGYtNGY3NC1hZTUzLWY0ZjIyYjY0NmQ4YS5naWY_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwMzI3JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDMyN1QxMzExMjNaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1hZjBjMGZjN2RiMGFiNDdlN2Y2Nzc1ZGZkY2FiZTc4Y2YwMzY5NDY5ZTY2NTY2YWU4ZmNjZTBhNmJhZmJlYTg5JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.SrsnflumboWC7_41SS_3d_d-VoPUTkDxCIB_Xyb53fA) [ ](https://private-user-images.githubusercontent.com/123875830/239668594-51e54c97-b20f-4f74-ae53-f4f22b646d8a.gif?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MTE1NDUzODMsIm5iZiI6MTcxMTU0NTA4MywicGF0aCI6Ii8xMjM4NzU4MzAvMjM5NjY4NTk0LTUxZTU0Yzk3LWIyMGYtNGY3NC1hZTUzLWY0ZjIyYjY0NmQ4YS5naWY_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwMzI3JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDMyN1QxMzExMjNaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1hZjBjMGZjN2RiMGFiNDdlN2Y2Nzc1ZGZkY2FiZTc4Y2YwMzY5NDY5ZTY2NTY2YWU4ZmNjZTBhNmJhZmJlYTg5JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.SrsnflumboWC7_41SS_3d_d-VoPUTkDxCIB_Xyb53fA) 你好,这个窗口滑动的功能是怎么实现的,可以发一下代码嘛,非常感谢。[email protected]

> > 如果方便的话,可以贴一下数据和代码我们看一下 > > 老师您好,数据我保存在百度网盘里了,链接:https://pan.baidu.com/s/1AsFctZe1FQIV8RoirWUsPw?pwd=nyja > > 提取码:nyja,tec_encoded.csv文件是栈式自编码器中间层输出的压缩结果,tec_features_encoded_2h_encoded_Mapminmax_normalized.csv文件里是对压缩数据进行了列归一化,将数据投影到(0, 1)以后的结果,将此文件里的数据输入模型训练,预测未来2018全年年4380个时间点数据,但预测结果的误差分布有较大问题,自己实验的误差分布形状类似于正太分布,而参考的IGS产品的误差分布为双驼峰形状,验证autoencoder的重构误差与IGS产品误差分布是一致的,观察预测结果发现好像整年的预测结果都差不多,感觉可能是这个问题,相应文件和代码都在百度网盘链接里面,烦请老师帮我看看是不是哪里还有问题,还有这样预测一年数据的方式是不是也有什么问题,谢谢老师 你好,这个链接过期了,能重新发一下嘛,我想学习学习,非常感谢

> 你好,我后来又仔细看了工程代码,发现我遗漏了很多重点;现在看来,在计算损失时只用到pred_len部分的数据,详见_process_one_batch函数中对batch_y的处理;所以与我上面说的数据拼接其实没关系,inverse的取值确实只和任务目标有关,想要得到更贴近原始标签的训练结果,还是需要把inverse设为True。我在个人的数据训练时也将inverse设为True,且通过调参也能获得较好的效果,建议多从数据预处理和参数选择入手试试 你好,麻烦问一下你都对哪些参数进行了调整吗?那几个参数对预测结果的影响比较大?谢谢了

你好,这个问题有什么进展嘛。