Sayan Mandal comments

Results 8 comments of


                                            Sayan Mandal

InvalidArgumentError (see above for traceback): indices[0,7] = -1 is not in [0, 157)

Solved it thanks to tensorflow gpu. Here is what I did: `conda create --name pixmesh tensorflow-gpu=1 python=2.7` then pip install tflearn

Problem

any solution on this? i got a similar error `2020-06-25 06:07:20.423197: W tensorflow/core/framework/op_kernel.cc:1651] OP_REQUIRES failed at gather_nd_op.cc:47 : Invalid argument: indices[1 17] = [2, 7] does not index into param...

Using `DistributedDataParallel` with PyG raises `TypeError: default_collate: batch must contain ...; found <class 'torch_geometric.data.data.Data'>`

``` from torch_geometric.datasets import Planetoid from torch.nn.parallel import DistributedDataParallel as DDP from torch.utils.data import DataLoader, Dataset import torch import os from floe.api import ComputeCube, DecimalParameter, BooleanParameter, IntegerParameter from orionplatform import...

Using `DistributedDataParallel` with PyG raises `TypeError: default_collate: batch must contain ...; found <class 'torch_geometric.data.data.Data'>`

Here is a snippet apologies for the delay. My only concern is loading the data in a multi GPU DDP setup. In this regard, this is where I am stuck:...

Using `DistributedDataParallel` with PyG raises `TypeError: default_collate: batch must contain ...; found <class 'torch_geometric.data.data.Data'>`

ahh I removed that line while trying to remove some commented out code. It should make sense now,using it right after I load the dataloader

Using `DistributedDataParallel` with PyG raises `TypeError: default_collate: batch must contain ...; found <class 'torch_geometric.data.data.Data'>`

let me try re-running, can you tell me the line number/reference line? I don't see `fds_to_keep` in the code. If its from the loadSingle_PytorchGeom function, I am not using it...

Using `DistributedDataParallel` with PyG raises `TypeError: default_collate: batch must contain ...; found <class 'torch_geometric.data.data.Data'>`

I am not using pytorch lightning for my code. I can if it is recommended. I was looking to optimize the dataloading part since i have a large amount of...

Using `DistributedDataParallel` with PyG raises `TypeError: default_collate: batch must contain ...; found <class 'torch_geometric.data.data.Data'>`

yes indeed! here are the relevant lines: ``` from torch_geometric.loader import DataLoader as GDL ptr = GDL(train_pair_graph, batch_size=batch_size, sampler=DistributedSampler(train_pair_graph)) ``` so far my train_pair_graph is a list. it works for...