Support for Pytorch DataLoader

[francescamanni1989]

Is it working only with LibSVMdata?

[sbadirli]

No, you can create your own dataloader as well.

…

On Mon, Jul 12, 2021 at 9:17 AM francescamanni1989 ***@***.***> wrote: Is it working only with LibSVMdata? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#6>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGTKHAZZFCYDQTYSHKUFKPLTXLTPPANCNFSM5AG2S42Q> .

[francescamanni1989]

when passing DataLoader like so:

I am getting the following error

I fixed previous error and now don't know which is the problem!

[sbadirli]

Interesting. It seems your x has 28 samples but somehow middle_feat (the result of the previous classifier ) has 16 samples in it. I believe your batch size is 16 right? right now I am at work but will try to look closely during the evening if I could squeeze some s-pare time for it.

…

On Mon, Jul 12, 2021 at 9:40 AM francescamanni1989 ***@***.***> wrote: when passing DataLoader like so: [image: image] <https://user-images.githubusercontent.com/83947886/125297224-58eba500-e327-11eb-9011-c8e7df684343.png> I am getting the following error [image: image] <https://user-images.githubusercontent.com/83947886/125297298-6acd4800-e327-11eb-9597-2450699d459f.png> I fixed previous error and now don't know which is the problem! — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGTKHA5L3JPCXSZ4EWJ5SZDTXLWE3ANCNFSM5AG2S42Q> .

[francescamanni1989]

Many thanks for your quick reply and support. When loading with your parameters for the MNIST usecase"

num_nets = 40
batch_size = 2048
#lr = [1.]
#L2 = [0.0001]
feat_d = 28
hidden_d = 16
sparse = False
epochs_per_stage = 1
correct_epoch=1
loss_models = torch.zeros((num_nets, 3))
model_order ='first'
all_ensm_losses = []
all_ensm_losses_te = []
all_mdl_losses = []
dynamic_br = []

boost_rate=1.
train_label=train_tensor_y.numpy()
c0 = np.mean(train_label)
net_ensemble = DynamicNet(c0, boost_rate)
I got the following. I am basically trying to load MNIST with train_loader but it is not working.

[sbadirli]

You are welcome. Are you sure MNIST has 28 features? With the current set-up, GrowNet works with only numeric data (NxM) (as we argued in the paper), thus feeding image (3D) will not work. It can definitely be extended to images which is my next project but I haven't had time for that yet. I am busy with my last year PhD and internship. Also GrowNet for now only supports binary classification, but it can be easily modified for multi-class classification.

…

On Mon, Jul 12, 2021 at 9:55 AM francescamanni1989 ***@***.***> wrote: Many thanks for your quick reply and support. When loading with your parameters for the MNIST usecase" num_nets = 40 batch_size = 2048 #lr = [1.] #L2 = [0.0001] feat_d = 28 hidden_d = 16 sparse = False epochs_per_stage = 1 correct_epoch=1 loss_models = torch.zeros((num_nets, 3)) model_order ='first' all_ensm_losses = [] all_ensm_losses_te = [] all_mdl_losses = [] dynamic_br = [] boost_rate=1. train_label=train_tensor_y.numpy() c0 = np.mean(train_label) net_ensemble = DynamicNet(c0, boost_rate) I got the following. I am basically trying to load MNIST with train_loader but it is not working. [image: image] <https://user-images.githubusercontent.com/83947886/125299444-728dec00-e329-11eb-836c-4a5ffdfb7254.png> — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGTKHAZSV4RSBFSSWM6O22DTXLX6VANCNFSM5AG2S42Q> .

[sbadirli]

Sorry for the disappointing answers. I will try to add these features within next year. Hope these answers help. On Mon, Jul 12, 2021 at 10:01 AM Sarkhan Badirli ***@***.***> wrote:

…

You are welcome. Are you sure MNIST has 28 features? With the current set-up, GrowNet works with only numeric data (NxM) (as we argued in the paper), thus feeding image (3D) will not work. It can definitely be extended to images which is my next project but I haven't had time for that yet. I am busy with my last year PhD and internship. Also GrowNet for now only supports binary classification, but it can be easily modified for multi-class classification. On Mon, Jul 12, 2021 at 9:55 AM francescamanni1989 < ***@***.***> wrote: > Many thanks for your quick reply and support. When loading with your > parameters for the MNIST usecase" > > num_nets = 40 > batch_size = 2048 > #lr = [1.] > #L2 = [0.0001] > feat_d = 28 > hidden_d = 16 > sparse = False > epochs_per_stage = 1 > correct_epoch=1 > loss_models = torch.zeros((num_nets, 3)) > model_order ='first' > all_ensm_losses = [] > all_ensm_losses_te = [] > all_mdl_losses = [] > dynamic_br = [] > > boost_rate=1. > train_label=train_tensor_y.numpy() > c0 = np.mean(train_label) > net_ensemble = DynamicNet(c0, boost_rate) > I got the following. I am basically trying to load MNIST with > train_loader but it is not working. > > [image: image] > <https://user-images.githubusercontent.com/83947886/125299444-728dec00-e329-11eb-836c-4a5ffdfb7254.png> > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#6 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AGTKHAZSV4RSBFSSWM6O22DTXLX6VANCNFSM5AG2S42Q> > . >

[francescamanni1989]

Thank you very much for your availability!