han-cai
I can only get the 88.3 +- acc, should I modify some param?
like the title, I can only get the 88.3 +- acc in validation, should I modify some param? such as ... epoch.... ???
hello, i get some trouble with this project. How to replace the "placeholder" with net.config? and how to create the ne.config? I'm confused about that, could you help me? thx
@sxs11 just run the run_densenet.py, then you can get the net.config... and you can move the net.config to your own "placeholder" dir...
Hey, may I ask why I got this error when I try to run arch_search.py with setting as densenet-cell? Have you ever encountered this error?
assert cell_architecture == Tree.build_tree_from_torch_module(net_config.building_block), 'Err' AssertionError: Err
Thanks a lot!
Hey, may I ask the accuracy you achieved? I have run the code with --setting=densenet-cell, but accuracy is only around 90.6% when max nets to train set as 300, which is much lower than the paper...
the best net i got when searching phase is around 90%。after trainning for 300 epoch， i got 93 the densenet used for 300 epoch is deeper than the one you used for searching，i think that is the reason 1692900722 邮箱：[email protected] 签名由 网易邮箱大师 定制 On 10/31/2018 15:13, noNameForMe1994 wrote: Hey, may I ask the accuracy you achieved? I have run the code with --setting=densenet-cell, but accuracy is only around 90.6% when max nets to train set as 300, which is much lower than the paper... — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.
the best net i got when searching phase is around 90%。after trainning for 300 epoch， i got 93 the densenet used for 300 epoch is deeper than the one you used for searching，i think that is the reason 1692900722 邮箱：[email protected] 签名由 网易邮箱大师 定制 On 10/31/2018 15:13, noNameForMe1994 wrote: Hey, may I ask the accuracy you achieved? I have run the code with --setting=densenet-cell, but accuracy is only around 90.6% when max nets to train set as 300, which is much lower than the paper... — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.
Thanks for that! But as you mentioned, the final accuracy achieved is still much lower than paper mentioned.. So I'm wondering if there are some tricks to train the network better?
is you net for trainging the same with the net obtained by searching？ in the paper， the net for training is different with the net which is obtained by searching 1692900722 邮箱：[email protected] 签名由 网易邮箱大师 定制 On 10/31/2018 17:50, noNameForMe1994 wrote: the best net i got when searching phase is around 90%。after trainning for 300 epoch， i got 93 the densenet used for 300 epoch is deeper than the one you used for searching，i think that is the reason 1692900722 邮箱：[email protected] 签名由 网易邮箱大师 定制 On 10/31/2018 15:13, noNameForMe1994 wrote: Hey, may I ask the accuracy you achieved? I have run the code with --setting=densenet-cell, but accuracy is only around 90.6% when max nets to train set as 300, which is much lower than the paper... — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread. Thanks for that! But as you mentioned, the final accuracy achieved is still much lower than paper mentioned.. So I'm wondering if there are some tricks to train the network better? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.
is you net for trainging the same with the net obtained by searching？ in the paper， the net for training is different with the net which is obtained by searching 1692900722 邮箱：[email protected] 签名由 网易邮箱大师 定制 On 10/31/2018 17:50, noNameForMe1994 wrote: the best net i got when searching phase is around 90%。after trainning for 300 epoch， i got 93 the densenet used for 300 epoch is deeper than the one you used for searching，i think that is the reason 1692900722 邮箱：[email protected] 签名由 网易邮箱大师 定制 On 10/31/2018 15:13, noNameForMe1994 wrote: Hey, may I ask the accuracy you achieved? I have run the code with --setting=densenet-cell, but accuracy is only around 90.6% when max nets to train set as 300, which is much lower than the paper... — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread. Thanks for that! But as you mentioned, the final accuracy achieved is still much lower than paper mentioned.. So I'm wondering if there are some tricks to train the network better? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.
I'll try that. Thank you so much!
@jiachen0212 @sxs11 I have run run_pyramidnet.py to get net.config for almost 10s... When I put net.config in my folder_dir and begin to arch_search.py, I got the Error raise JSONDecodeError("Expecting value", s, err.value) from None json.decoder.JSONDecodeError: Expecting value: line 2 column 25 (char 26) There must be sth I got wrong. Could you tell me how to search arc with the codes? THX!
