Codazzy
Codazzy
Hi @ShenchenYu98, 我们近期对cogdl进行了较大更新。麻烦试一下升级到cogdl最新版本（比如安装最新的master分支，或者安装0.5.0.a1版本）。然后按照[自定义数据集教程](https://cogdl.readthedocs.io/en/latest/tutorial/custom_dataset.html#dataset-for-graph-classification)中的示例用法来实现试试。
Hi @YuanSQ1997, 感谢你对我们工作的关注！你可以查阅一下我们的文档：https://cogdl.readthedocs.io/en/latest/tutorial/custom_dataset.html。
Hi @huang429, Thanks for your interest in cogdl. Could you please report your running command or script?
Hi @huihuijiangqiang, 感谢对cogdl的关注！请问你是想获取模型最后预测出来的结果（tensor）吗？
可以参考这个自定义模型的例子：https://github.com/THUDM/cogdl/blob/master/examples/custom_gcn.py 把外面定义model传进experiment API。训练完成后直接用这个model对数据集做预测。
Hi @Cedric-Kram, Thanks for your attention to cogdl. We will take a look at this bug ASAP.
Hi @Cedric-Kram, In graph2vec model, the features (formed as a tensor) will be first fed into `feature_extractor` and transformed into strings. I wonder what exactly your problem is. Do you...
Hi @odymit, Thanks for your interest in our toolkit. From your description, you have 5000 graphs for graph classification. Each graph has 512(?) nodes with 512-dim(?) features. Therefore, it will...
Hi @odymit, Currently, we assume that all data need to be loaded into the memory at the beginning of the training. Therefore, CogDL only supports the dataset where the size...
Hi @IrisLi17, Good job. But I think maybe it's better to implement DropEdge GCN on an independent model file rather than the original GCN file.