比如不能用pip再安装一些其他的库，如果我用的算法库环境里没有，还得自己手动写一遍？
Posted by: wizcheu @ Nov. 27, 2019, 11:06 a.m.If you want to install additional packages, please use python os module to execute "pip install XX" commands at the beginning of your model.py file.
Posted by: Zhen @ Nov. 28, 2019, 5:04 a.m.Will the installation time of needed python packages be included into the running time of the program?
Posted by: lishuqiao @ Dec. 2, 2019, 8:55 a.m.Will the installation time of needed python packages be included into the running time of the program?
Posted by: lishuqiao @ Dec. 2, 2019, 8:55 a.m.No, this time will be excluded.
Posted by: Zhen @ Dec. 2, 2019, 9:17 a.m.@Zhen, our SUBMISSION ID 22913 failed, with package torch installation time ~= 2 minute and the score.py program failed our submission before the installation done. Is it normal ?
Ingestion LOG:
2019-12-03 11:05:19,696 INFO ingestion.py: ===== import user model
Collecting torch
Downloading https://files.pythonhosted.org/packages/88/95/90e8c4c31cfc67248bf944ba42029295b77159982f532c5689bcfe4e9108/torch-1.3.1-cp36-cp36m-manylinux1_x86_64.whl (734.6MB)
Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from torch) (1.17.2)
Installing collected packages: torch
Successfully installed torch-1.3.1
info: {'is_multivariate': True, 'is_relative_time': False, 'label': 'A29', 'primary_id': ['A28'], 'primary_timestamp': 'A1', 'schema': {'A1': 'timestamp', 'A10': 'num', 'A11': 'num', 'A12': 'num', 'A13': 'num', 'A14': 'num', 'A15': 'num', 'A16': 'num', 'A17': 'num', 'A18': 'num', 'A19': 'num', 'A2': 'num', 'A20': 'num', 'A21': 'num', 'A22': 'num', 'A23': 'num', 'A24': 'num', 'A25': 'num', 'A26': 'num', 'A27': 'num', 'A28': 'str', 'A29': 'num', 'A3': 'num', 'A4': 'num', 'A5': 'num', 'A6': 'num', 'A7': 'num', 'A8': 'num', 'A9': 'num'}, 'time_budget': {'load': 10.0, 'predict': 1000.0, 'save': 10.0, 'train': 100.0, 'update': 200.0}}
sample of test record: 7706
number of pred timestamp: 3853
Finish init
2019-12-03 11:07:46,124 INFO ingestion.py: ===== Begin training user model
Score.py LOG:
2019-12-03 11:05:19,889 DEBUG score.py: --------------------------------------------------
2019-12-03 11:05:19,889 DEBUG score.py: Using solution_dir: /tmp/codalab/hidden_ref_89ed18ed-07d1-4081-99c9-579d6a71352c
2019-12-03 11:05:19,889 DEBUG score.py: Using prediction_dir: /tmp/codalab/tmpXbsOUs/run/input/res
2019-12-03 11:05:19,889 DEBUG score.py: Using score_dir: /tmp/codalab/tmpXbsOUs/run/output
2019-12-03 11:05:19,889 INFO score.py: ===== wait for ingestion to start
2019-12-03 11:05:19,890 INFO score.py: ===== detect alive ingestion
2019-12-03 11:05:19,890 INFO score.py: ===== wait for the exit of ingestion or end.yaml file
2019-12-03 11:06:36,468 INFO score.py: no alive ingestion
2019-12-03 11:06:36,469 ERROR score.py: no end.yaml exist, ingestion failed
It's a scoring program issue and we have already fixed it, you should be able to install additional packages now, please try it again. Sorry for the Inconvenient.
Posted by: liushouxiang @ Dec. 11, 2019, 12:34 p.m.平台环境导致的包依赖问题，比如更新pandas这个包，因为在model.py中有save/load操作，即使在文件最顶部更新pandas(比import pandas还要提前)，而第一次save到本地的pickle文件中包含的也是旧版本的pandas，会导致一些异常，是否有可能提供一个conda环境？而且这样安装依赖包的方式并不是很友好，各种包之间的依赖问题很难解决，很多时间浪费在包依赖问题上，找不到完美的解决方案。
Posted by: wizcheu @ Dec. 11, 2019, 2:40 p.m.今天numpy的版本默认升级到1.17.2了？下次环境做了改动能否在主页公告一下。。。作为基础库，它的版本会影响一些其他库的使用= =
Posted by: wizcheu @ Dec. 12, 2019, 10:24 p.m.Dear participants,
We didn't upgrade the version of numpy, also we double checked the version on the competition server, it's still 1.15.1.(Actually, we provide the libraries in the docker image "vergilgxw/autotable:v3", the version of all libraries should be stably since we didn't upgrade the image).
So how do you find that the version come out to 1.17.2? More details would be helpful. Thanks.
AutoSeries Competition Organizers
Posted by: liushouxiang @ Dec. 19, 2019, 8:45 a.m.