中级农民
- 积分
- 216
- 大米
- 颗
- 鳄梨
- 个
- 水井
- 尺
- 蓝莓
- 颗
- 萝卜
- 根
- 小米
- 粒
- 学分
- 个
- 注册时间
- 2013-8-26
- 最后登录
- 1970-1-1
|
High Level:
At the end of the day, it is comparing A vs. B. Which one is better?
Choosing and characterizing metrics:
Define Metrics: valuation, revenue, experience?
- 1. high level active users
- 2. details, what is active
- 3. decide the logic
Refine:
- 1. funnel, basically different steps of the process
- 2. rate, % from one step to next
- 3. click through rate/prob: Rates are often better than probabilities for measuring the usability of a button, and increasing the size of the "Start Now" button is probably an attempt to increase the usability. Thus, click-through-rate might have been a better choice. However, click-through-probability is a good choice as well. Rate is by event, prob is by users.
- 4. difficult metrics: external data, existing data, focus group, survey, UER- User experience research (show correlation, not causation, weakness of retrospective analysis)
- 5. take into account of company's culture, focus
Filter and segmenting:
- calc on different slices
- build intuitions
- try diff time intervals
- talk to engineers if find unexpected
Summary Metrics:
- sensitive & robust
- 4 categories
- sensitivity: pick changes your care about
- robust: ignore changes you don't care about
- mean or median? mean sensitive to outliers, not robust; if only a fraction, median not as robust, may add percentile
- benefit of hindsight
- how to measure S&R: latency example, experiments or retrospective. by diff videos, 90th percentile not very robust (comprable videos, similar resolution)
https://towardsdatascience.com/a ... course-9ecc32dedbb1
|
|