b*****o
发帖数: 715 | 1 Machine Learning:
The High-Interest Credit Card of Technical Debt
http://static.googleusercontent.com/media/research.google.com/e
其中我很感慨的就是下面这段：自己重写一个专用的工具比开源的通用工具靠谱。
It may be surprising to the academic community to know that only a tiny
fraction of the code in
many machine learning systems is actually doing “machine learning”. When
we recognize that a
mature system might end up being (at most) 5% machine learning code and (at
least) 95% glue code,
reimplementation rather than reuse of a clumsy API looks like a much better
strategy.
我自己加的推论就是，production的ML系统，千万不能用python来写，虽然刚上手的时
候很爽，但是太难维护和扩展了。 |
d******e
发帖数: 7844 | 2 有用Python来做产品的？
at
【在 b*****o 的大作中提到】
: Machine Learning:
: The High-Interest Credit Card of Technical Debt
: http://static.googleusercontent.com/media/research.google.com/e
: 其中我很感慨的就是下面这段：自己重写一个专用的工具比开源的通用工具靠谱。
: It may be surprising to the academic community to know that only a tiny
: fraction of the code in
: many machine learning systems is actually doing “machine learning”. When
: we recognize that a
: mature system might end up being (at most) 5% machine learning code and (at
: least) 95% glue code,
|
Z**0
发帖数: 1119 | 3 Oh my god. 95% glue code.
要么是里边有很多machine learning module，需要glue together。更加可能是，采集
的数据，需要很多步骤处理，才可以用在ML的module上面，最后还有presentation
result部分。
是不是大部分DS在公司很多工作都是和clean data和present final result有关了？
at
【在 b*****o 的大作中提到】
: Machine Learning:
: The High-Interest Credit Card of Technical Debt
: http://static.googleusercontent.com/media/research.google.com/e
: 其中我很感慨的就是下面这段：自己重写一个专用的工具比开源的通用工具靠谱。
: It may be surprising to the academic community to know that only a tiny
: fraction of the code in
: many machine learning systems is actually doing “machine learning”. When
: we recognize that a
: mature system might end up being (at most) 5% machine learning code and (at
: least) 95% glue code,
|
b*****o
发帖数: 715 | 4 Machine Learning:
The High-Interest Credit Card of Technical Debt
http://static.googleusercontent.com/media/research.google.com/e
其中我很感慨的就是下面这段：自己重写一个专用的工具比开源的通用工具靠谱。
It may be surprising to the academic community to know that only a tiny
fraction of the code in
many machine learning systems is actually doing “machine learning”. When
we recognize that a
mature system might end up being (at most) 5% machine learning code and (at
least) 95% glue code,
reimplementation rather than reuse of a clumsy API looks like a much better
strategy.
我自己加的推论就是，production的ML系统，千万不能用python来写，虽然刚上手的时
候很爽，但是太难维护和扩展了。 |
d******e
发帖数: 7844 | 5 有用Python来做产品的？
at
【在 b*****o 的大作中提到】
: Machine Learning:
: The High-Interest Credit Card of Technical Debt
: http://static.googleusercontent.com/media/research.google.com/e
: 其中我很感慨的就是下面这段：自己重写一个专用的工具比开源的通用工具靠谱。
: It may be surprising to the academic community to know that only a tiny
: fraction of the code in
: many machine learning systems is actually doing “machine learning”. When
: we recognize that a
: mature system might end up being (at most) 5% machine learning code and (at
: least) 95% glue code,
|
Z**0
发帖数: 1119 | 6 Oh my god. 95% glue code.
要么是里边有很多machine learning module，需要glue together。更加可能是，采集
的数据，需要很多步骤处理，才可以用在ML的module上面，最后还有presentation
result部分。
是不是大部分DS在公司很多工作都是和clean data和present final result有关了？
at
【在 b*****o 的大作中提到】
: Machine Learning:
: The High-Interest Credit Card of Technical Debt
: http://static.googleusercontent.com/media/research.google.com/e
: 其中我很感慨的就是下面这段：自己重写一个专用的工具比开源的通用工具靠谱。
: It may be surprising to the academic community to know that only a tiny
: fraction of the code in
: many machine learning systems is actually doing “machine learning”. When
: we recognize that a
: mature system might end up being (at most) 5% machine learning code and (at
: least) 95% glue code,
|
l********e
发帖数: 220 | 7 为啥不能用python?看不出来你这个推论怎么来的？不用python 用啥？C++, R?有什么
区别？R的ML API也很多把?
at
【在 b*****o 的大作中提到】
: Machine Learning:
: The High-Interest Credit Card of Technical Debt
: http://static.googleusercontent.com/media/research.google.com/e
: 其中我很感慨的就是下面这段：自己重写一个专用的工具比开源的通用工具靠谱。
: It may be surprising to the academic community to know that only a tiny
: fraction of the code in
: many machine learning systems is actually doing “machine learning”. When
: we recognize that a
: mature system might end up being (at most) 5% machine learning code and (at
: least) 95% glue code,
|
