ZexuanSun / DTF-Drug-Synergy

Codes and data for DTF
4 stars 3 forks source link

您好!麻烦请教一个问题。 #1

Closed jamessmith123456 closed 4 years ago

jamessmith123456 commented 4 years ago

请问项目文件夹下的./data_sets/final_data/drug_a.csv, drug_b.csv和cell_line.csv是怎么得来的呢?是直接运行cpwopt_final.m得来的吗? cpwopt_final.m中,将fold_0_final.csv中的项在P矩阵中设置为0? 我运行cpwopt_final.m得到的特征值,跟./data_sets/final_data/drug_a.csv, drug_b.csv和fold_0_final.csv中的不一样;想不出为啥,能不能请教您一下,感谢!

ZexuanSun commented 4 years ago

他这个东西有一定的随机性,因为CP-WOPT那个优化算法每次结果都不一样,他的目标函数只是逼近原来的tensor,逼近的可能有很多种。 所以我觉得这东西也有点离谱,这样再加一个DNN,又加了一个随机性,这样预测最后跑出来的结果也有随机性,但是有些pair是经常出现在最后的结果里的我记得。😑😑😑综上我觉得这个model的稳定性就有点离谱。见笑了不好意思hhhh

Best, Zexuan


发件人: jamessmith123456 notifications@github.com 发送时间: 星期四, 六月 11, 2020 09:06 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Subscribed 主题: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

请问项目文件夹下的./data_sets/final_data/drug_a.csv, drug_b.csv和cell_line.csv是怎么得来的呢?是直接运行cpwopt_final.m得来的吗? cpwopt_final.m中,将fold_0_final.csv中的项在P矩阵中设置为0? 我运行cpwopt_final.m得到的特征值,跟./data_sets/final_data/drug_a.csv, drug_b.csv和fold_0_final.csv中的不一样;想不出为啥,能不能请教您一下,感谢!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJF75R3JYQ37I4V5O6HP2SDRWAUYRANCNFSM4N26NOZA.

jamessmith123456 commented 4 years ago

非常感谢您的回复! (1) 是的,我多次运行时发现每次CP-WOPT的结果都不一样,存在随机性; (2) 另外我想请教的问题是,文章中的结果部分,Accuracy、Precision这些指标值是5-fold得来的吗?还是在测试集上(也就是120239个真正的缺失项,非validation data)得来的? 我在数据集中没找到缺失项对应的synergy得分,以及它们对应的0-1标签; (3)最后请教一个私人的问题,实在抱歉打扰您,您不方便的话可以不回复,感谢!我想请问这个课题还有继续研究的价值吗,似乎没有太多关于drug-drug-synergy方向的论文出现...

再次感谢您的回复!您的回复帮我解决了问题,谢谢您!

------------------ 原始邮件 ------------------ 发件人: "Zexuan Sun"<notifications@github.com>; 发送时间: 2020年6月11日(星期四) 上午9:18 收件人: "ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com>; 抄送: "庄驰"<1321985517@qq.com>;"State change"<state_change@noreply.github.com>; 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

他这个东西有一定的随机性,因为CP-WOPT那个优化算法每次结果都不一样,他的目标函数只是逼近原来的tensor,逼近的可能有很多种。 所以我觉得这东西也有点离谱,这样再加一个DNN,又加了一个随机性,这样预测最后跑出来的结果也有随机性,但是有些pair是经常出现在最后的结果里的我记得。😑😑😑综上我觉得这个model的稳定性就有点离谱。见笑了不好意思hhhh

Best, Zexuan


发件人: jamessmith123456 <notifications@github.com> 发送时间: 星期四, 六月 11, 2020 09:06 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Subscribed 主题: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

请问项目文件夹下的./data_sets/final_data/drug_a.csv, drug_b.csv和cell_line.csv是怎么得来的呢?是直接运行cpwopt_final.m得来的吗? cpwopt_final.m中,将fold_0_final.csv中的项在P矩阵中设置为0? 我运行cpwopt_final.m得到的特征值,跟./data_sets/final_data/drug_a.csv, drug_b.csv和fold_0_final.csv中的不一样;想不出为啥,能不能请教您一下,感谢!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1&gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R3JYQ37I4V5O6HP2SDRWAUYRANCNFSM4N26NOZA&gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

ZexuanSun commented 4 years ago

(2)是5 fold的,不是真正的缺失项,真正缺失项不知道实际值有些metrics没法算,你看的是我arxiv上的draft吧,我那边没有更新到最后accept的版本有些地方没写清楚,然后这里每次算出来也是有随机性的,因为有个dnn,但稳定性还算好我记得 (3)这个我也不晓得啊,我老板当时给我的是这个课题,我就做了这个,我觉得那个只是应用背景在drug synergy,可以变成tensor的背景还是挺多的,那个model的思想其实我感觉可以用在别的地方,如果你指的是drug synergy这个方向那我真的不太清楚,你是做生信的吗,这块我也不晓得,不知道bioinformtics什么比较火,我之前老板dl ml这块用的挺多,但是他后来还叫他别的学生用tensor做别的东西,他之前还让我看transformer什么的说这是dl最新的东西,我没看😂😂😂

PS: 不用这么客气,用“你”就好了😂😂😂


发件人: jamessmith123456 notifications@github.com 发送时间: 星期四, 六月 11, 2020 20:19 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Zexuan Sun; Comment 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

非常感谢您的回复! (1) 是的,我多次运行时发现每次CP-WOPT的结果都不一样,存在随机性; (2) 另外我想请教的问题是,文章中的结果部分,Accuracy、Precision这些指标值是5-fold得来的吗?还是在测试集上(也就是120239个真正的缺失项,非validation data)得来的? 我在数据集中没找到缺失项对应的synergy得分,以及它们对应的0-1标签; (3)最后请教一个私人的问题,实在抱歉打扰您,您不方便的话可以不回复,感谢!我想请问这个课题还有继续研究的价值吗,似乎没有太多关于drug-drug-synergy方向的论文出现...

再次感谢您的回复!您的回复帮我解决了问题,谢谢您!

------------------ 原始邮件 ------------------ 发件人: "Zexuan Sun"<notifications@github.com>; 发送时间: 2020年6月11日(星期四) 上午9:18 收件人: "ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com>; 抄送: "庄驰"<1321985517@qq.com>;"State change"<state_change@noreply.github.com>; 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

他这个东西有一定的随机性,因为CP-WOPT那个优化算法每次结果都不一样,他的目标函数只是逼近原来的tensor,逼近的可能有很多种。 所以我觉得这东西也有点离谱,这样再加一个DNN,又加了一个随机性,这样预测最后跑出来的结果也有随机性,但是有些pair是经常出现在最后的结果里的我记得。😑😑😑综上我觉得这个model的稳定性就有点离谱。见笑了不好意思hhhh

Best, Zexuan


发件人: jamessmith123456 <notifications@github.com> 发送时间: 星期四, 六月 11, 2020 09:06 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Subscribed 主题: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

请问项目文件夹下的./data_sets/final_data/drug_a.csv, drug_b.csv和cell_line.csv是怎么得来的呢?是直接运行cpwopt_final.m得来的吗? cpwopt_final.m中,将fold_0_final.csv中的项在P矩阵中设置为0? 我运行cpwopt_final.m得到的特征值,跟./data_sets/final_data/drug_a.csv, drug_b.csv和fold_0_final.csv中的不一样;想不出为啥,能不能请教您一下,感谢!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1&gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R3JYQ37I4V5O6HP2SDRWAUYRANCNFSM4N26NOZA&gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1#issuecomment-642607570, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJF75R2YC4WHXPUJTE7TPJLRWDDUXANCNFSM4N26NOZA.

jamessmith123456 commented 4 years ago

嘿嘿,老哥,感谢你的回复!你给我点一下,我的疑惑就解答了! 我目前也是刚接触这个方向,所以对tensor以及drug synergy的理解不深,感觉不知从何下手,所以想着向大佬请教下; 老哥,我暂时问题就这么些,非常感谢你的解答和建议, 我再去看看有没有别的文章;后面有问题我再来叨扰老哥!

老哥晚安!加油!

------------------ 原始邮件 ------------------ 发件人: "Zexuan Sun"<notifications@github.com>; 发送时间: 2020年6月11日(星期四) 晚上11:02 收件人: "ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com>; 抄送: "庄驰"<1321985517@qq.com>;"State change"<state_change@noreply.github.com>; 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

(2)是5 fold的,不是真正的缺失项,真正缺失项不知道实际值有些metrics没法算,你看的是我arxiv上的draft吧,我那边没有更新到最后accept的版本有些地方没写清楚,然后这里每次算出来也是有随机性的,因为有个dnn,但稳定性还算好我记得 (3)这个我也不晓得啊,我老板当时给我的是这个课题,我就做了这个,我觉得那个只是应用背景在drug synergy,可以变成tensor的背景还是挺多的,那个model的思想其实我感觉可以用在别的地方,如果你指的是drug synergy这个方向那我真的不太清楚,你是做生信的吗,这块我也不晓得,不知道bioinformtics什么比较火,我之前老板dl ml这块用的挺多,但是他后来还叫他别的学生用tensor做别的东西,他之前还让我看transformer什么的说这是dl最新的东西,我没看😂😂😂

PS: 不用这么客气,用“你”就好了😂😂😂


发件人: jamessmith123456 <notifications@github.com> 发送时间: 星期四, 六月 11, 2020 20:19 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Zexuan Sun; Comment 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

非常感谢您的回复! (1) 是的,我多次运行时发现每次CP-WOPT的结果都不一样,存在随机性; (2) 另外我想请教的问题是,文章中的结果部分,Accuracy、Precision这些指标值是5-fold得来的吗?还是在测试集上(也就是120239个真正的缺失项,非validation data)得来的? 我在数据集中没找到缺失项对应的synergy得分,以及它们对应的0-1标签; (3)最后请教一个私人的问题,实在抱歉打扰您,您不方便的话可以不回复,感谢!我想请问这个课题还有继续研究的价值吗,似乎没有太多关于drug-drug-synergy方向的论文出现...

再次感谢您的回复!您的回复帮我解决了问题,谢谢您!

------------------&nbsp;原始邮件&nbsp;------------------ 发件人:&nbsp;"Zexuan Sun"<notifications@github.com&gt;; 发送时间:&nbsp;2020年6月11日(星期四) 上午9:18 收件人:&nbsp;"ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com&gt;; 抄送:&nbsp;"庄驰"<1321985517@qq.com&gt;;"State change"<state_change@noreply.github.com&gt;; 主题:&nbsp;Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

他这个东西有一定的随机性,因为CP-WOPT那个优化算法每次结果都不一样,他的目标函数只是逼近原来的tensor,逼近的可能有很多种。 所以我觉得这东西也有点离谱,这样再加一个DNN,又加了一个随机性,这样预测最后跑出来的结果也有随机性,但是有些pair是经常出现在最后的结果里的我记得。😑😑😑综上我觉得这个model的稳定性就有点离谱。见笑了不好意思hhhh

Best, Zexuan


发件人: jamessmith123456 <notifications@github.com&gt; 发送时间: 星期四, 六月 11, 2020 09:06 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Subscribed 主题: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

请问项目文件夹下的./data_sets/final_data/drug_a.csv, drug_b.csv和cell_line.csv是怎么得来的呢?是直接运行cpwopt_final.m得来的吗? cpwopt_final.m中,将fold_0_final.csv中的项在P矩阵中设置为0? 我运行cpwopt_final.m得到的特征值,跟./data_sets/final_data/drug_a.csv, drug_b.csv和fold_0_final.csv中的不一样;想不出为啥,能不能请教您一下,感谢!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1&amp;gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R3JYQ37I4V5O6HP2SDRWAUYRANCNFSM4N26NOZA&amp;gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1#issuecomment-642607570&gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R2YC4WHXPUJTE7TPJLRWDDUXANCNFSM4N26NOZA&gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

ZexuanSun commented 4 years ago

好的好的,欢迎随时交流,晚安老哥😴


发件人: jamessmith123456 notifications@github.com 发送时间: Thursday, June 11, 2020 11:10:04 PM 收件人: ZexuanSun/DTF-Drug-Synergy DTF-Drug-Synergy@noreply.github.com 抄送: Zexuan Sun zexuansun@outlook.com; Comment comment@noreply.github.com 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

嘿嘿,老哥,感谢你的回复!你给我点一下,我的疑惑就解答了! 我目前也是刚接触这个方向,所以对tensor以及drug synergy的理解不深,感觉不知从何下手,所以想着向大佬请教下; 老哥,我暂时问题就这么些,非常感谢你的解答和建议, 我再去看看有没有别的文章;后面有问题我再来叨扰老哥!

老哥晚安!加油!

------------------ 原始邮件 ------------------ 发件人: "Zexuan Sun"<notifications@github.com>; 发送时间: 2020年6月11日(星期四) 晚上11:02 收件人: "ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com>; 抄送: "庄驰"<1321985517@qq.com>;"State change"<state_change@noreply.github.com>; 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

(2)是5 fold的,不是真正的缺失项,真正缺失项不知道实际值有些metrics没法算,你看的是我arxiv上的draft吧,我那边没有更新到最后accept的版本有些地方没写清楚,然后这里每次算出来也是有随机性的,因为有个dnn,但稳定性还算好我记得 (3)这个我也不晓得啊,我老板当时给我的是这个课题,我就做了这个,我觉得那个只是应用背景在drug synergy,可以变成tensor的背景还是挺多的,那个model的思想其实我感觉可以用在别的地方,如果你指的是drug synergy这个方向那我真的不太清楚,你是做生信的吗,这块我也不晓得,不知道bioinformtics什么比较火,我之前老板dl ml这块用的挺多,但是他后来还叫他别的学生用tensor做别的东西,他之前还让我看transformer什么的说这是dl最新的东西,我没看😂😂😂

PS: 不用这么客气,用“你”就好了😂😂😂


发件人: jamessmith123456 <notifications@github.com> 发送时间: 星期四, 六月 11, 2020 20:19 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Zexuan Sun; Comment 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

非常感谢您的回复! (1) 是的,我多次运行时发现每次CP-WOPT的结果都不一样,存在随机性; (2) 另外我想请教的问题是,文章中的结果部分,Accuracy、Precision这些指标值是5-fold得来的吗?还是在测试集上(也就是120239个真正的缺失项,非validation data)得来的? 我在数据集中没找到缺失项对应的synergy得分,以及它们对应的0-1标签; (3)最后请教一个私人的问题,实在抱歉打扰您,您不方便的话可以不回复,感谢!我想请问这个课题还有继续研究的价值吗,似乎没有太多关于drug-drug-synergy方向的论文出现...

再次感谢您的回复!您的回复帮我解决了问题,谢谢您!

------------------&nbsp;原始邮件&nbsp;------------------ 发件人:&nbsp;"Zexuan Sun"<notifications@github.com&gt;; 发送时间:&nbsp;2020年6月11日(星期四) 上午9:18 收件人:&nbsp;"ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com&gt;; 抄送:&nbsp;"庄驰"<1321985517@qq.com&gt;;"State change"<state_change@noreply.github.com&gt;; 主题:&nbsp;Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

他这个东西有一定的随机性,因为CP-WOPT那个优化算法每次结果都不一样,他的目标函数只是逼近原来的tensor,逼近的可能有很多种。 所以我觉得这东西也有点离谱,这样再加一个DNN,又加了一个随机性,这样预测最后跑出来的结果也有随机性,但是有些pair是经常出现在最后的结果里的我记得。😑😑😑综上我觉得这个model的稳定性就有点离谱。见笑了不好意思hhhh

Best, Zexuan


发件人: jamessmith123456 <notifications@github.com&gt; 发送时间: 星期四, 六月 11, 2020 09:06 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Subscribed 主题: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

请问项目文件夹下的./data_sets/final_data/drug_a.csv, drug_b.csv和cell_line.csv是怎么得来的呢?是直接运行cpwopt_final.m得来的吗? cpwopt_final.m中,将fold_0_final.csv中的项在P矩阵中设置为0? 我运行cpwopt_final.m得到的特征值,跟./data_sets/final_data/drug_a.csv, drug_b.csv和fold_0_final.csv中的不一样;想不出为啥,能不能请教您一下,感谢!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1&amp;gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R3JYQ37I4V5O6HP2SDRWAUYRANCNFSM4N26NOZA&amp;gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1#issuecomment-642607570&gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R2YC4WHXPUJTE7TPJLRWDDUXANCNFSM4N26NOZA&gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1#issuecomment-642724887, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJF75R7GVIBC7CDBOCBLEBDRWDXUZANCNFSM4N26NOZA.

jamessmith123456 commented 4 years ago

老哥老哥,是我呀,我之前向你请教问题的😁😁 老哥,你那个论文里和DeepSynergy这个方法进行了对比,效果比它略好些;我看DeepSynergy这个文章,它自己放出来的结果和老哥你对比用到的结果不一样;它自己的值高一些; 请问老哥,这个是为啥呀?我看数据集是同一份数据集呀,老哥的那个DeepSynergy得分是怎么来的呢? 麻烦老哥啦!嘻嘻~感谢感谢~

---原始邮件--- 发件人: "Zexuan Sun"<notifications@github.com> 发送时间: 2020年6月11日(周四) 晚上11:12 收件人: "ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com>; 抄送: "State change"<state_change@noreply.github.com>;"jamessmith123456"<1321985517@qq.com>; 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

好的好的,欢迎随时交流,晚安老哥😴


发件人: jamessmith123456 <notifications@github.com> 发送时间: Thursday, June 11, 2020 11:10:04 PM 收件人: ZexuanSun/DTF-Drug-Synergy <DTF-Drug-Synergy@noreply.github.com> 抄送: Zexuan Sun <zexuansun@outlook.com>; Comment <comment@noreply.github.com> 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

嘿嘿,老哥,感谢你的回复!你给我点一下,我的疑惑就解答了! 我目前也是刚接触这个方向,所以对tensor以及drug synergy的理解不深,感觉不知从何下手,所以想着向大佬请教下; 老哥,我暂时问题就这么些,非常感谢你的解答和建议, 我再去看看有没有别的文章;后面有问题我再来叨扰老哥!

老哥晚安!加油!

------------------&nbsp;原始邮件&nbsp;------------------ 发件人:&nbsp;"Zexuan Sun"<notifications@github.com&gt;; 发送时间:&nbsp;2020年6月11日(星期四) 晚上11:02 收件人:&nbsp;"ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com&gt;; 抄送:&nbsp;"庄驰"<1321985517@qq.com&gt;;"State change"<state_change@noreply.github.com&gt;; 主题:&nbsp;Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

(2)是5 fold的,不是真正的缺失项,真正缺失项不知道实际值有些metrics没法算,你看的是我arxiv上的draft吧,我那边没有更新到最后accept的版本有些地方没写清楚,然后这里每次算出来也是有随机性的,因为有个dnn,但稳定性还算好我记得 (3)这个我也不晓得啊,我老板当时给我的是这个课题,我就做了这个,我觉得那个只是应用背景在drug synergy,可以变成tensor的背景还是挺多的,那个model的思想其实我感觉可以用在别的地方,如果你指的是drug synergy这个方向那我真的不太清楚,你是做生信的吗,这块我也不晓得,不知道bioinformtics什么比较火,我之前老板dl ml这块用的挺多,但是他后来还叫他别的学生用tensor做别的东西,他之前还让我看transformer什么的说这是dl最新的东西,我没看😂😂😂

PS: 不用这么客气,用“你”就好了😂😂😂


发件人: jamessmith123456 <notifications@github.com&gt; 发送时间: 星期四, 六月 11, 2020 20:19 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Zexuan Sun; Comment 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

非常感谢您的回复! (1) 是的,我多次运行时发现每次CP-WOPT的结果都不一样,存在随机性; (2) 另外我想请教的问题是,文章中的结果部分,Accuracy、Precision这些指标值是5-fold得来的吗?还是在测试集上(也就是120239个真正的缺失项,非validation data)得来的? 我在数据集中没找到缺失项对应的synergy得分,以及它们对应的0-1标签; (3)最后请教一个私人的问题,实在抱歉打扰您,您不方便的话可以不回复,感谢!我想请问这个课题还有继续研究的价值吗,似乎没有太多关于drug-drug-synergy方向的论文出现...

再次感谢您的回复!您的回复帮我解决了问题,谢谢您!

------------------&amp;nbsp;原始邮件&amp;nbsp;------------------ 发件人:&amp;nbsp;"Zexuan Sun"<notifications@github.com&amp;gt;; 发送时间:&amp;nbsp;2020年6月11日(星期四) 上午9:18 收件人:&amp;nbsp;"ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com&amp;gt;; 抄送:&amp;nbsp;"庄驰"<1321985517@qq.com&amp;gt;;"State change"<state_change@noreply.github.com&amp;gt;; 主题:&amp;nbsp;Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

他这个东西有一定的随机性,因为CP-WOPT那个优化算法每次结果都不一样,他的目标函数只是逼近原来的tensor,逼近的可能有很多种。 所以我觉得这东西也有点离谱,这样再加一个DNN,又加了一个随机性,这样预测最后跑出来的结果也有随机性,但是有些pair是经常出现在最后的结果里的我记得。😑😑😑综上我觉得这个model的稳定性就有点离谱。见笑了不好意思hhhh

Best, Zexuan


发件人: jamessmith123456 <notifications@github.com&amp;gt; 发送时间: 星期四, 六月 11, 2020 09:06 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Subscribed 主题: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

请问项目文件夹下的./data_sets/final_data/drug_a.csv, drug_b.csv和cell_line.csv是怎么得来的呢?是直接运行cpwopt_final.m得来的吗? cpwopt_final.m中,将fold_0_final.csv中的项在P矩阵中设置为0? 我运行cpwopt_final.m得到的特征值,跟./data_sets/final_data/drug_a.csv, drug_b.csv和fold_0_final.csv中的不一样;想不出为啥,能不能请教您一下,感谢!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1&amp;amp;gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R3JYQ37I4V5O6HP2SDRWAUYRANCNFSM4N26NOZA&amp;amp;gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1#issuecomment-642607570&amp;gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R2YC4WHXPUJTE7TPJLRWDDUXANCNFSM4N26NOZA&amp;gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1#issuecomment-642724887&gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R7GVIBC7CDBOCBLEBDRWDXUZANCNFSM4N26NOZA&gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

ZexuanSun commented 4 years ago

1.我重新跑了一遍,deepsynergy那个也是一样的DNN跑出来结果每次都不一样的,但是差不太多 2.deepsynergy跟DTF结果是similar也没有谁好谁坏吧,最后的版本里面做了t-test,给出的结论是similar,arxiv上我那个不是终稿,我一直没更新....


发件人: jamessmith123456 notifications@github.com 发送时间: Monday, July 6, 2020 1:46:30 AM 收件人: ZexuanSun/DTF-Drug-Synergy DTF-Drug-Synergy@noreply.github.com 抄送: Zexuan Sun zexuansun@outlook.com; Comment comment@noreply.github.com 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

老哥老哥,是我呀,我之前向你请教问题的😁😁 老哥,你那个论文里和DeepSynergy这个方法进行了对比,效果比它略好些;我看DeepSynergy这个文章,它自己放出来的结果和老哥你对比用到的结果不一样;它自己的值高一些; 请问老哥,这个是为啥呀?我看数据集是同一份数据集呀,老哥的那个DeepSynergy得分是怎么来的呢? 麻烦老哥啦!嘻嘻~感谢感谢~

---原始邮件--- 发件人: "Zexuan Sun"<notifications@github.com> 发送时间: 2020年6月11日(周四) 晚上11:12 收件人: "ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com>; 抄送: "State change"<state_change@noreply.github.com>;"jamessmith123456"<1321985517@qq.com>; 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

好的好的,欢迎随时交流,晚安老哥😴


发件人: jamessmith123456 <notifications@github.com> 发送时间: Thursday, June 11, 2020 11:10:04 PM 收件人: ZexuanSun/DTF-Drug-Synergy <DTF-Drug-Synergy@noreply.github.com> 抄送: Zexuan Sun <zexuansun@outlook.com>; Comment <comment@noreply.github.com> 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

嘿嘿,老哥,感谢你的回复!你给我点一下,我的疑惑就解答了! 我目前也是刚接触这个方向,所以对tensor以及drug synergy的理解不深,感觉不知从何下手,所以想着向大佬请教下; 老哥,我暂时问题就这么些,非常感谢你的解答和建议, 我再去看看有没有别的文章;后面有问题我再来叨扰老哥!

老哥晚安!加油!

------------------&nbsp;原始邮件&nbsp;------------------ 发件人:&nbsp;"Zexuan Sun"<notifications@github.com&gt;; 发送时间:&nbsp;2020年6月11日(星期四) 晚上11:02 收件人:&nbsp;"ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com&gt;; 抄送:&nbsp;"庄驰"<1321985517@qq.com&gt;;"State change"<state_change@noreply.github.com&gt;; 主题:&nbsp;Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

(2)是5 fold的,不是真正的缺失项,真正缺失项不知道实际值有些metrics没法算,你看的是我arxiv上的draft吧,我那边没有更新到最后accept的版本有些地方没写清楚,然后这里每次算出来也是有随机性的,因为有个dnn,但稳定性还算好我记得 (3)这个我也不晓得啊,我老板当时给我的是这个课题,我就做了这个,我觉得那个只是应用背景在drug synergy,可以变成tensor的背景还是挺多的,那个model的思想其实我感觉可以用在别的地方,如果你指的是drug synergy这个方向那我真的不太清楚,你是做生信的吗,这块我也不晓得,不知道bioinformtics什么比较火,我之前老板dl ml这块用的挺多,但是他后来还叫他别的学生用tensor做别的东西,他之前还让我看transformer什么的说这是dl最新的东西,我没看😂😂😂

PS: 不用这么客气,用“你”就好了😂😂😂


发件人: jamessmith123456 <notifications@github.com&gt; 发送时间: 星期四, 六月 11, 2020 20:19 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Zexuan Sun; Comment 主题: Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

非常感谢您的回复! (1) 是的,我多次运行时发现每次CP-WOPT的结果都不一样,存在随机性; (2) 另外我想请教的问题是,文章中的结果部分,Accuracy、Precision这些指标值是5-fold得来的吗?还是在测试集上(也就是120239个真正的缺失项,非validation data)得来的? 我在数据集中没找到缺失项对应的synergy得分,以及它们对应的0-1标签; (3)最后请教一个私人的问题,实在抱歉打扰您,您不方便的话可以不回复,感谢!我想请问这个课题还有继续研究的价值吗,似乎没有太多关于drug-drug-synergy方向的论文出现...

再次感谢您的回复!您的回复帮我解决了问题,谢谢您!

------------------&amp;nbsp;原始邮件&amp;nbsp;------------------ 发件人:&amp;nbsp;"Zexuan Sun"<notifications@github.com&amp;gt;; 发送时间:&amp;nbsp;2020年6月11日(星期四) 上午9:18 收件人:&amp;nbsp;"ZexuanSun/DTF-Drug-Synergy"<DTF-Drug-Synergy@noreply.github.com&amp;gt;; 抄送:&amp;nbsp;"庄驰"<1321985517@qq.com&amp;gt;;"State change"<state_change@noreply.github.com&amp;gt;; 主题:&amp;nbsp;Re: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

他这个东西有一定的随机性,因为CP-WOPT那个优化算法每次结果都不一样,他的目标函数只是逼近原来的tensor,逼近的可能有很多种。 所以我觉得这东西也有点离谱,这样再加一个DNN,又加了一个随机性,这样预测最后跑出来的结果也有随机性,但是有些pair是经常出现在最后的结果里的我记得。😑😑😑综上我觉得这个model的稳定性就有点离谱。见笑了不好意思hhhh

Best, Zexuan


发件人: jamessmith123456 <notifications@github.com&amp;gt; 发送时间: 星期四, 六月 11, 2020 09:06 收件人: ZexuanSun/DTF-Drug-Synergy 抄送: Subscribed 主题: [ZexuanSun/DTF-Drug-Synergy] 您好!麻烦请教一个问题。 (#1)

请问项目文件夹下的./data_sets/final_data/drug_a.csv, drug_b.csv和cell_line.csv是怎么得来的呢?是直接运行cpwopt_final.m得来的吗? cpwopt_final.m中,将fold_0_final.csv中的项在P矩阵中设置为0? 我运行cpwopt_final.m得到的特征值,跟./data_sets/final_data/drug_a.csv, drug_b.csv和fold_0_final.csv中的不一样;想不出为啥,能不能请教您一下,感谢!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1&amp;amp;gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R3JYQ37I4V5O6HP2SDRWAUYRANCNFSM4N26NOZA&amp;amp;gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1#issuecomment-642607570&amp;gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R2YC4WHXPUJTE7TPJLRWDDUXANCNFSM4N26NOZA&amp;gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub<https://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1#issuecomment-642724887&gt;, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJF75R7GVIBC7CDBOCBLEBDRWDXUZANCNFSM4N26NOZA&gt;.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/ZexuanSun/DTF-Drug-Synergy/issues/1#issuecomment-653917912, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJF75RZO2FYLIRSSPBJZF3TR2C37NANCNFSM4N26NOZA.