昨天小群上的科学家提及到ML,今天看了一个视频非常简单易懂。做一个笔记方便以后继续学习。今天还看到了一个论调说好多最近兴起的topic其实也没有那么的难。入门级别应该花点功夫就能达到。以此也提醒自己,随时随地抱着好奇心去关注自己觉得有兴趣的事物吧。
ML回答的5个问题(就只有5个,没其他了。然后会有相对的Algorithm去对付这些问题):
1. Is this A or B (or C...): Classification algorithm, handle multi-class questions, handle question with a number of answers
2. Is this weird: Anomaly detection algorithm
3. How much/How many: Regression Algorithm, any question that asks for a number
4. How is this organized: Clustering Algorithm: no one right answer, but help organize structure and better predict behavior/event
5. What should I do now: Reinforcement Learning, make a lot of small decision without human guidance
显然: 4和5是open ended question,所以对ML要求更高了
Data Science:
Algorithm=Recipe
Data=Ingredients
Computer=Blender
Anser=Smoothie
下一步是要针对data是不是好的。 有5个原则:Relevant (data类型之间有没有联系), connected(就算类型相关,有没有missing呢), accurate (是否否会指导错误的结论), enough to work with (不够的话结论很fuzzy)
想Data问问题很重要. 一定要问Sharp question. 并且考虑问问题的角度, 和目前是不是已经有target data在data base了 (例如问下星期stock price的价格, 首先要有stock的历史价格)
网友评论