search

anonymousgit2020 / fedsem

BSD 2-Clause "Simplified" License
26 stars 5 forks source link

About the arguments of model #1

Open meiyuan666 opened 4 years ago

meiyuan666 commented 4 years ago

Hi , I am glad to read your paper, Multi-Center Federated Learning. The work in the paper is innovative and interesting,from which I get a lot of inspiration. However, when I reproduce your experiment, for example: run python fedcluster.py, I do not know how to choose the arguments of "model". Could you give me some more guidance? Thank you very much!

sky1106 commented 4 years ago

Hi , I am glad to read your paper, Multi-Center Federated Learning. The work in the paper is innovative and interesting,from which I get a lot of inspiration. However, when I reproduce your experiment, for example: run python fedcluster.py, I do not know how to choose the arguments of "model". Could you give me some more guidance? Thank you very much!

Hi, I meet the same problem with you, if you have found the solution? Could you please give me an example of all arguments to run the fedcluster.py? Thank you every much

meiyuan666 commented 4 years ago

Hi , I am glad to read your paper, Multi-Center Federated Learning. The work in the paper is innovative and interesting,from which I get a lot of inspiration. However, when I reproduce your experiment, for example: run python fedcluster.py, I do not know how to choose the arguments of "model". Could you give me some more guidance? Thank you very much!

Hi, I meet the same problem with you, if you have found the solution? Could you please give me an example of all arguments to run the fedcluster.py? Thank you every much

I haven't found a solution yet, sorry~

sky1106 commented 4 years ago

Hi , I am glad to read your paper, Multi-Center Federated Learning. The work in the paper is innovative and interesting,from which I get a lot of inspiration. However, when I reproduce your experiment, for example: run python fedcluster.py, I do not know how to choose the arguments of "model". Could you give me some more guidance? Thank you very much! Hi, I meet the same problem with you, if you have found the solution? Could you please give me an example of all arguments to run the fedcluster.py? Thank you every much

I haven't found a solution yet, sorry~

OK~ thank you all the same, we can wait for the author's reply

moFang222 commented 4 years ago

@meiyuan666 @sky1106 Hi! Did you guys reproduce the result of FedSM algorithm in the original paper?I can't achieve its accuracy when reproducing, I wonder if this is the cause of some certain hyperparameters, because the paper does not specify these hyperparameters in detail.

sky1106 commented 4 years ago

@meiyuan666 @sky1106 Hi! Did you guys reproduce the result of FedSM algorithm in the original paper?I can't achieve its accuracy when reproducing, I wonder if this is the cause of some certain hyperparameters, because the paper does not specify these hyperparameters in detail.

Hi! You have already reproduced the algorithm? Could you please tell me the arguments that you ran with? We can disscuss about this. By the way, did you find some bugs when you reproduced the algorithm?

moFang222 commented 4 years ago

@sky1106 If you mean the FedSME algorithm,I ran it with code python3 mlhead.py -dataset femnist -model cnn --clients-per-round 20 --num-epochs 5 -lr 0.003.I haven't run the whole experiment yet, but the accuracy is almost 0.1 when I reach 30 rounds

sky1106 commented 4 years ago

@sky1106 If you mean the FedSME algorithm,I ran it with code python3 mlhead.py -dataset femnist -model cnn --clients-per-round 20 --num-epochs 5 -lr 0.003.I haven't run the whole experiment yet, but the accuracy is almost 0.1 when I reach 30 rounds

It works not for me, there are lots of bugs to be fixed. You did not meet these bugs when you reproduced the algorithm? bugs

moFang222 commented 4 years ago

@sky1106 It looks like something went wrong with the femnist data set. Did you prepare the data set as described here ？

sky1106 commented 4 years ago

@sky1106 It looks like something went wrong with the femnist data set. Did you prepare the data set as described here ？

OK~I still meet lots of bugs, I will fix them first and then tell you my results. Thank you very much!

moFang222 commented 4 years ago

@sky1106 ok!And please try some different hyperparameters than those I listed above to make sure you find the hyperparameters that produce the best result!Waiting for your result！

meiyuan666 commented 4 years ago

@moFang222 @sky1106 Thank you for all your replies! And have you already reproduced the experiment of FedCluster?

moFang222 commented 4 years ago

@moFang222 @sky1106 Thank you for all your replies! And have you already reproduced the experiment of FedCluster?

@meiyuan666 Sorry, I didn't see FedCluster in the paper, do you mean FedSEM algorithm?Or the Clustered baseline algorithm in the red box below. Snipaste_2020-06-08_19-55-05

meiyuan666 commented 4 years ago

@moFang222 @sky1106 Thank you for all your replies! And have you already reproduced the experiment of FedCluster?

@meiyuan666 Sorry, I didn't see FedCluster in the paper, do you mean FedSEM algorithm?Or the Clustered baseline algorithm in the red box below. Snipaste_2020-06-08_19-55-05

Yeah, the baseline algorithm, which is proposed by the paper : Clustered federated learning: Model-agnostic distributed multi-task optimization under privacy constraints. The relevant implementation code is also in this repository: all_baselines/fed-cluster/models/fedcluster.py.

moFang222 commented 4 years ago

@meiyuan666 Sorry,I have not reproduced this algorithm yet.

meiyuan666 commented 4 years ago

@meiyuan666 Sorry,I have not reproduced this algorithm yet.

It doesn't matter.Thank you very much for your reply~

peiji1981 commented 4 years ago

hi guys, I am also doing research about federated clustering algorithms. I wanna ask for a paper/code/idea about vertically federated clustering, did you see that ? thx you all ~