Fw: Bug on MetaDL Track1 online server

[超宇 关]

Dear Organizers:

Any update on this bug? Our code suffers this issue badly 🙁 We think maybe the (cpu) memory cost is too large for tensorflow to load data so that the process is killed by the system. If so, maybe some changes need done in the environment code (metadl and meta-dataset library).

Also, we notice some bugs on time counting of online platform. Our time left for dataset 4 today run out only after we make one submission. The platform says that Execution Time Left: -1 day, 18:48:10, which is so strange. Can you help fix this?

Sincerely,

Meta_Learners

2021.09.17

---

From: 超宇 关
Sent: Friday, September 10, 2021 21:25
To: metalearni...@googlegroups.com <metalearni...@googlegroups.com>
Subject: Bug on MetaDL Track1 online server

 

Dear organizers:

We are team Meta_Learners and think there is some bug on the online server of MetaDL Track1. We've made a submission (ID: 44722) which just read the data from given dataset and print the time repeatedly.

We've put the "read and print" code into an endless loop. However, according to the output log, the program vanishes after only 5 minutes.

Here is the log:

current time - epoch  1  -  2021-09-10 10:53:57
current time - epoch  2  -  2021-09-10 10:54:14
current time - epoch  3  -  2021-09-10 10:54:31
current time - epoch  4  -  2021-09-10 10:54:48
current time - epoch  5  -  2021-09-10 10:55:04
current time - epoch  6  -  2021-09-10 10:55:21
current time - epoch  7  -  2021-09-10 10:55:37
current time - epoch  8  -  2021-09-10 10:55:53
current time - epoch  9  -  2021-09-10 10:56:10
current time - epoch  10  -  2021-09-10 10:56:26
current time - epoch  11  -  2021-09-10 10:56:43
current time - epoch  12  -  2021-09-10 10:56:59
current time - epoch  13  -  2021-09-10 10:57:16
current time - epoch  14  -  2021-09-10 10:57:32
current time - epoch  15  -  2021-09-10 10:57:48
current time - epoch  16  -  2021-09-10 10:58:05

The time is definitely enough because the code starts at 

2021-09-10 10:53:25

which can be inferred from the log of warning message. Can you help check whether the program vanishes online? Are there some bugs related to processing the dataset? Many thanks!

Sincerely,

Meta_Learners

2021.09.10