Gemini icon indicating copy to clipboard operation
Gemini copied to clipboard

step and acc

Open zzjss12 opened this issue 1 year ago • 15 comments

  1. May I ask how to adjust the following three values when the data volume changes? step_ per_ epoch,valid_ step_ per_ epoch,test_ step_ per_ epoch

  2. What is the reason why the ACC value has been around 50%?

zzjss12 avatar May 06 '23 09:05 zzjss12

image 这里面的 graph_embedding = tf.reduce_sum(curr_embedding, axis=2) curr_embedding 这个变量应该是prev_embedding吧

zzjss12 avatar Jun 05 '23 12:06 zzjss12

image The threshold here should be 0.5. The problem has been resolved.

zzjss12 avatar Jun 06 '23 12:06 zzjss12

  1. May I ask how to adjust the following three values when the data volume changes? step_ per_ epoch,valid_ step_ per_ epoch,test_ step_ per_ epoch
  2. What is the reason why the ACC value has been around 50%?

请问你成功复现了吗,可以交流一下吗

Hululuhh avatar Jan 06 '24 08:01 Hululuhh

@Hululuhh 请问你遇到同样的问题了吗?

Yunlongs avatar Jan 09 '24 08:01 Yunlongs

@Hululuhh 请问你遇到同样的问题了吗?

对的,我还发现使用您的数据训练过程中valid的auc一直为0,导致没有模型保存,这是为什么?感谢您,如果方便的话,[email protected]是我的邮箱,我希望可以得到您的帮助

Hululuhh avatar Jan 09 '24 08:01 Hululuhh

@Hululuhh 你这边方便把zzjss12同学所说的threshold改为0.5然后再看下问题是否存在吗?这有可能是我之前更新代码产生的bug,谢谢

Yunlongs avatar Jan 09 '24 09:01 Yunlongs

好的,我会修改一下再告诉你结果的,希望最近可以跟您保持联系

------------------ 原始邮件 ------------------ 发件人: "Yunlongs/Gemini" @.>; 发送时间: 2024年1月9日(星期二) 下午5:54 @.>; @.@.>; 主题: Re: [Yunlongs/Gemini] step and acc (Issue #5)

@Hululuhh 你这边方便把zzjss12同学所说的threshold改为0.5然后再看下问题是否存在吗?这有可能是我之前更新代码产生的bug,谢谢

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

Hululuhh avatar Jan 09 '24 09:01 Hululuhh

@Hululuhh 你这边方便把zzjss12同学所说的threshold改为0.5然后再看下问题是否存在吗?这有可能是我之前更新代码产生的bug,谢谢

我修改了,看起来train的Accuracy会到75左右,但valid和test还是40,并且auc为0

Hululuhh avatar Jan 10 '24 03:01 Hululuhh

@Hululuhh 那看起来threshold的值是0.5时训练才是正确的。AUC值为0,可能是代码里的计算方式有问题

Yunlongs avatar Jan 11 '24 02:01 Yunlongs

@Hululuhh 那看起来threshold的值是0.5时训练才是正确的。AUC值为0,可能是代码里的计算方式有问题

image valid代码中加上这行好像就可以了

iDTer avatar Apr 16 '24 03:04 iDTer

哈喽各位大佬,请问你们是怎么解决保存模型时报错

ValueError: Model <__main__.MyModel object at 0x000001CD174F7D88> cannot be saved because the input shapes have not been set. Usually, input shapes are automatically determined from calling .fit() or .predict(). To manually set the shapes, call model._set_inputs(inputs).

的问题的呀

qrzbing avatar Apr 24 '24 15:04 qrzbing