This entry rookie machine learning there for some time, so there is a Diudiu sentiment, here to do a little summary. Tell us about the learning experience of machine learning theory and practice.
Related materials
Mathematical basis
High number of linear algebra which nothing to say, is a compulsory subject in the University of Engineering.
Statistical Machine Learning
Li Hang's Blue Book and Zhou Zhihua of watermelon book can be said is the more classic textbook, the two also in the field of artificial intelligence, the domestic leader.
Depth study
Strongly recommended flower book , it can be said that the authoritative textbook learning in depth. In addition there is Andrew Ng lectures and instructional videos, can be found online.
White introductory textbooks
Recommended earlier books, although regarded as entry materials, but may be for white, it is not very easy to accept. It is recommended that entry-depth study: Based on the theory and implementation of Python and learn practical machine two books, speak easy to understand, easy to use.
Advanced materials
PRML and MLAPP these books belong to God book level, to see who knows who.
additional materials
Direct search machine learning in depth study on github or keywords, you can find a lot of high-Star project, I will not list them.
Practical part
- Recommend programming language python, simple and easy to use. Now also launched a Chinese official document: https://docs.python.org/zh-cn/3/
- Machine learning, python, there are three very important package: numpy, pandas, matplot. Use of specific methods consult the appropriate official document: https://www.numpy.org/devdocs/
https://pandas.pydata.org/pandas-docs/stable/ https://matplotlib.org/index.html
these three packages being no official Chinese, but there are some folk Chinese version, you can own Baidu. - sklearn, this package integrates almost all statistical machine learning API, is a very important tool bag. Official Website: http://scikit-learn.github.io/stable
- pytorch, a deep learning framework, of course, and now a lot of deep learning framework, such as: tensorflow, keras and so on. But pytorch is relatively easy to use. Official Address: https://pytorch.org/
- Some packages are based on the specific tasks to be used, for example, do nlp often need to use nltk, jieba and other packages, this case I will not go into detail.
paper
Things teaching materials are relatively lag, academic front or on paper, machine learning this piece upgrading particularly fast, we still have much more to read the latest articles to learn about the latest algorithms and models.
Sentiment
Do machine learning, many times, and in the transfer package parameter adjustment, although we do Tucao, but in fact a lot of people will not even transfer package. Using these packages, we need to spend a lot of time to read the document, familiar API parameters , so as to write better programs. In addition, we also need to read the paper, so little innovation on the model.