Fan Hong comments

Results 11 comments of


                                            Fan Hong

alink webui 中拖动算子出现“实验数据建立中，请稍后报错”

Please refer to discussion in this issue #204

Throws below error when i tried to submit Flink job to session cluster. any help would be appreciated

It seems the root cause is > Caused by: com.typesafe.config.ConfigException$Missing: No configuration setting found for key 'silvertail' And from the last several lines, it seems the exception comes from `com.cdo.fraud.Job.main`.

alink1.5.2 getOutputTable().to_pandas() raise ERROR

Hi. `to_pandas` is provided by PyFlink, not PyAlink. From the error message, it seems `to_pandas` does not support Alink-defined types: `: java.lang.UnsupportedOperationException: Python vectorized UDF doesn't support logical type LEGACY('RAW',...

请问有计划实现卷积神经网络、LSTM等算法吗？

Alink 现在支持了与 TF 脚本的对接，包括分布式训练的功能，也能够支持各种深度模型的推理功能。可以参考这里的文档：https://www.yuque.com/pinshu/alink_tutorial/gyybst

Stuck when calling System.load on Windows

Does anyone has an idea? I googled and found some similar cases: https://stackoverflow.com/questions/5262552/loadlibrary-freezes https://forums.codeguru.com/showthread.php?244147-LoadLibrary-blocking It seems reading blocking from stdin in another thread makes loadLibrary hang.

How does this compare to Dask?

Hi, I am just curious about this: "What is interesting that the Graph in Mars can support loop, we may implement some loop semantic in the future. I think it...

Poor performance of sequence parallel fused kernels in real model training

I have two conjectures now: 1. The CPU code is just slow, because there are some calls of tensor `split` and `view`. 2. There is a global lock in the...

Poor performance of sequence parallel fused kernels in real model training

Thanks for your kind replies, @danthe3rd and @lw . @lw I've tried triton kernel. But in my case, the seq length is not fixed. Then for almost every batch of...

Poor performance of sequence parallel fused kernels in real model training

Yeah, I've already tried padding and is acceptable in our training.