“Live up to the time, the creation of non-stop, this article is participating in 2021 year-end summary essay competition”

preface

Fleeting, blink of an eye 2021 has entered the end, this is an ordinary also not ordinary year. Generally compared with previous years, there is no difference, nothing more than busy work, a good meal; What’s unusual is that there are still so many unforgettable things to remember this year.

Reading a book

Overall, I didn’t read too many books this year, and most of them were non-technical.

I still remember seeing teacher Luo Xiang’s short video about reading on Douyin, which gave me a new understanding of reading. He said, read not to show off, read only to develop your wisdom. We are often good at reading without understanding. We don’t need to memorize the wonderful passages, we just need to read them, and gradually they become our nourishment. We need to speak to the great soul of mankind.

I agree with a sentence: reading is to open the window of the heart, try to read 20 books next year.

The reading list is as follows:

  • Silicon Valley Iron Man
  • How to read a book
  • Kubernetes In Action
  • Betty’s Colors
  • Critical dialogue
  • 7. Everyone is a Product Manager (Completion: 20%)
  • Python machine learning (Level of Completion: 50%)
  • Low-risk entrepreneurship
  • Value – Zhang Lei
  • China cloud native AI white paper in Taiwan
  • Redis Combat (Completion: 20%)

writing

I have written 57 blogs this year, including 55 technical ones and 2 non-technical ones. In general, the creation of this year’s blog or quite satisfied, of course, thanks to nuggets, this July found nuggets this technical community, and then in nuggets collected a lot of wool, as follows:

  1. Participated in the August Genwen Challenge, completed 30 articles in total and passed the fourth level.
  2. Participated in the creation activity of “Essential Tips for Programmers” and completed 16 pieces in total, reaching the level of brick miners.
  3. Participated in the November Genwen Challenge, completed 30 articles in total and passed the fourth level.

By participating in nuggets activities, I was also encouraged to constantly learn some new knowledge.

The corresponding writing list is as follows:

Technical – 55

Artificial Intelligence Engineering (MLOps) -10

  • Open source machine learning workflow Ploomber
  • Analyze data workflow Prefect
  • Machine Learning Kubflow Pipelines
  • Analysis of Automatic machine learning (AutoML) tool NNI
  • Artificial Intelligence Systems (I) : Overview
  • Artificial Intelligence Systems (II) : Technology Stack
  • Open Source Workflow Tools Survey (MLOps)
  • 12 elements of Repeatable and reproducible machine learning in production
  • Forward data Engineering and Reverse Data Engineering (Prefect workflows)
  • A global overview of MLOps and ML tools

Artificial Intelligence practice -21

  • Master the evaluation index of classification algorithm in ten minutes
  • Master the evaluation index of regression algorithm in 10 minutes
  • Ten minutes to master the evaluation index of clustering algorithm
  • Text feature extraction method in SkLearn
  • Sklearn is used for feature selection
  • Machine learning visualization tool -Yellowbrick
  • Data missing value visualization tool – Missingno
  • Analysis of data preprocessing method in SKLearn
  • Analysis of three sequence labeling methods of named Entity Recognition (NER)
  • Model evaluation indicators in multi-label classification scenarios
  • Common model evaluation indicators for multi-label classification scenarios in SKLearn
  • Some methods of text and image data enhancement are briefly described
  • Some solutions to data imbalance are discussed briefly
  • Analysis of Pipeline in SKlearn
  • Data preprocessing for discrete features: independent thermal coding, label coding and binarization
  • Overview of skLearn model evaluation index functions for different classification scenarios
  • Several methods of discretization of feature data are analyzed
  • Classification model evaluation index of SKLearn (I) : accuracy rate, Top accuracy rate, balance accuracy rate
  • Evaluation indexes of classification model in SKLearn (II) : Kappa coefficient, confusion matrix, classification index report, hamming loss
  • Classification model evaluation index of SKLearn (III) : accuracy rate, recall rate, F value
  • Evaluation indexes of classification model in SKLearn (IV) : Jaccard similarity coefficient, hinge loss and logarithmic loss

Python Backend Technology Practices -4

  • Python dependency management and packaging tool -Poetry
  • An overview of Python processes, threads, and coroutines
  • Learn how to communicate between processes in Python
  • Brief description, create objects, and view data

Container and Cloud Practices -7

  • Analysis of Kubernetes controller
  • Docker resource (CPU/ memory/Disk IO/GPU) restriction and allocation guide
  • Pure dry! Thirteen best practice points for building Dockfile images
  • Pure dry! How to elegantly simplify Docker images?
  • Docker 中 国 内 容
  • Pure dry! Docker Dockerfile command
  • Kubernetes 解 决 方 法

Middleware Technology -5

It mainly involves message queue (Kafka), search engine (ElasticSearch), database (mysql), cache (Redis).

  • Deploying ElasticSearch on Kubernetes via ECK How to add custom username and password?
  • Illustrate the difference between cache breakdown, cache penetration, and cache avalanche
  • A brief Guide to Elasticsearch Development Specification (Continuously updated)
  • MySQL Tuning Guide (Continuously updated)
  • MySQL database SQL usage specifications

Mathematics and Computer Science -5

  • Discussion on P, NP, NP-complate and NP-hard problems
  • Analysis of five IO models in Linux
  • Read the common bandwidth in computer systems
  • This section describes common disk performance indicators and commands for observing I/O performance indicators
  • Understand the difference between IO intensive and CPU intensive in a minute

Other – 3

  • The reason why the jobs could not be viewed after the jobs command was executed was finally found
  • PyTorch training deep learning model in Docker or Kubernetes with insufficient shared memory
  • Modify the author, email, and name information in Git commit history

Non-technical category -2

  • Reading Notes – Key dialogues
  • Describe the intelligent conversation system

A sideline to make money

After working for a few years, I feel more and more that creating a side business is very graduate. First, it was able to restructure its revenue model (from a single wage to multiple channels) to better deal with potential risks. Second, it can reach out to different circles and see more future possibilities. At the same time, it can bring more positive changes to our body and mind, and bring more confidence.

Current sideline income for financial management income and creative income.

Financial income

Financial income mainly through reasonable buy fund, in fact, graduated from 2016 to now, I have to buy fund, early, just bought is where to put, did not spend too much time to study (of course the most began to also buy not much), until the end of last year began to research foundation, understand the industry, understand the fund manager’s past performance, As well as pondering their own investment strategy, investment fund mentality also slowly become mature.

In general, the returns of the fund this year are very average, not outperforming GDP as a whole, but better than csi 300 in the same period. The specific reasons are analyzed as follows:

  1. The whole environment of A shares is not ideal (liquor, new energy, medical and other industries rose too much last year, this year in the digestion of high valuation).
  2. In June this year, a large increase in the position to buy the Internet, thinking that can copy the bottom, the result of copying in the middle of the mountain, the purchase of the Internet fund overall loss of about 25%.

Create income

At present, the creation income comes from Zhihu. At present, the creation income is mainly obtained through the recommendation of good things and the payment of knowledge. This income is quite unexpected this year, because I did not expect to earn money through creation. I don’t earn much, but I’m still happy.

research

Last year, due to the organizational structure adjustment of the department, I changed my job from intelligent customer service to AI center. The whole tech stack has changed, whether it’s Docker, Kubernete or AI, it’s not my area of expertise. In fact, not only I, the entire department is relatively ignorant of this part of the technology stack. Therefore, to introduce some relevant technologies, we need to do the corresponding technical research. In addition, for products like AI Zhongtai, it is difficult to do a good job by completely relying on product managers to drive product design. Therefore, we also need to drive product design from a technical perspective. Therefore, we will also do some product research and competitive product analysis.

Technology research

This year, I mainly conducted technical research in two directions, one is machine learning assembly line and the other is data version management. These two pieces are part of the MLOps.

The corresponding technical survey list is as follows:

Machine learning pipeline

  • Airflow
  • Prefect
  • ZenML
  • Kubenetes Pipeline
  • Ploomber

Data version Management

  • DVC
  • Git LFS

Product research

This year, I mainly conducted product research on intelligent dialogue engine and AI medium platform (NLP platform and OCR platform).

The corresponding product research list is as follows:

  • Baidu UNIT
  • Baidu EasyDL
  • Ali NLP platform

conclusion

All in all, I learned a lot this year. Thanks to the time, growth happened unexpectedly, and 2022 came as expected. Come on, cook! Heaven will bless those who brave the storm.