大二结束哩
我跌跌撞撞连滚带爬被2022抽进7月,这学期过的用颠沛流离形容也不为过,哥们心里那是一个五味杂陈😥。
考完试了,弄个复盘,也是极为重要的。
这学期从2月开始上网课,一直拉到7月,捏麻麻地,体验感极差,哥们2-3月处于鬼混打游戏时期,4月返校惨遭内卷毒打,开始复习,5月有一段真空期过的还算是舒适,6月从1号到30号,全程自闭,压力拉满。
这学期有个最大的感受是,没有目标,或者说目标太大不到咋去做,总感觉有种无头苍蝇的感觉,因为要做的东西一股脑呈现在面前,又被卷地不得不去做。有句话是这样说的,“不要因为别人交了答卷,自己就开始随便填答案”,zxr这学期算是完美践行了这句话的反例。
这学期的几门主课基本都没咋花功夫细学,然后主要搞的那个nlp也没弄好,综合评价为哈哈。
但是说起来总体的进步还是有的,从1月的还在看mit 6.034到现在撸模型,算是囫囵吞枣大概糊了一个框架,但是基础太差,我又太菜,就也做出来啥东西,挫败感肯定是有,加上马上大三要准备研究生了,就压力也是跟上来了。
我也是越来越感到自己菜的离谱,也不是贬低自己,算是认清现状了,学的那些个东西都很浅而且不完善,就内种你给别人3个 ...
nlp learning recording
nlp learning recording
General Procedure Of Preprocessing
Normalization
Tokenization
Stop words
Part-of-Speech Tagging
Named Entity Recognition
Stemming and Lemmatization
Word Embedding
One-hot Vector
Description: most simple way to represent categorical feature one-hot.
Disadvantages:
For large vocabularies, these vectors will be very long: vector dimensionality is equal to the vocabulary size.
These vectors know nothing about the words they represent, one-hot vectors do not capture m ...
随便记记
dc9fcc9c819d5a4f43fe8e7daa086cf61d64277f53779768f95dc77b0be8b39c105c36c28322a40f87bfe34ec821ccecc219cb76a9025884217b0a0b8e279771b769e25a971178ef2a5d16899edf242680612c2755385a04067006ecee72b0fa0fff1fd9ba04063a2c9c1978181574b2c6bcaf664b9d85736ee7ecf85607fa874deaed5179630046de71072342c4779fb25c0e609f7f2baabd82d484424e9b8fa90336d862330641b27870f7ab24059ba4cfdebc408d7e6e39ff1f785a38fbd6b215190e5c4d58cbf1e0e81e044d342f74e562ff0a71857a326c99c2313e3e42dd888e3026b4a227b5609a699ff78cad004804731f9e6b794 ...
Multimodal NER
概述&原模型
写在开始
论文代码仓库:https://github.com/softhuafei/Pytorch-implementation-for-OCSGA
其实是少了object_label这个文件,需要翻历史记录下载,我也不到为啥把这个删了。
这应该是我第一次改论文代码,对于整体的代码框架啥的其实都不是很清楚,还有具体用到的一些方法,也不是特别熟,所以就来边学边改,做个记录。
----2022/5/30 update----
和博士师兄讨论完,感觉还是任重道远,bug确实不少,要弄的东西也蛮多的,所以既然这玩意不是只靠查查资料就能解决的,还是要静下心来,研读代码,毕竟这个代码也是挺规范的,还有很多地方值得学习,所以还是要耐住躁动和寂寞🚴♂️。
----2022/6/5 update----
哥们现在过拟合的问题还是没解决,感概良多,但是还是要踏踏实实地把这个东西做完,感觉心情失落是预期过高,对任务的分析不够客观,所以更需要静下心来调整分析。虽然说每天就答辩了,感觉具体的结果还是应该延期到交论文的时候提交,继续努力!
----2022/6/18 upd ...
摸鱼学方法论
还是写个小复盘(摸鱼方法论)吧,这学期基本一直也没写过这种小总结,应该是我太懒了。
刚考完计算方法,还是发现自己有不少问题的,整个应对考试的机制居然和高中差不多,乐。我一直认为,如果get stuck on only one task,基本上是永远没法做好的,因为整个体系没有留白,效率也会跟不上。正确的做法应该是分配好时间,一定是要上手多个task,而且要制定好阶段性目标,个人感觉不要太难,重点是坚持,对于某一项近期十分捉急的task,把权重调高一些,但是也绝不是dominant地位(除非明天就要考试了,溜)。
应试
先说说应试吧,考试来说的话,有种感觉就是复习完了以后,结果反而有种自己也说不清的不信任,感觉做题也会有些疑虑,这种可能是考试前2天才开始复习刷题(指zxr),导致的刷题还没形成肌肉记忆,这就是平时只去记背后的intuition和推导过程,而完全妹有刷题的恶果,现在回想起来,反而是考完试的第二天对于这门学科的体系印象更深刻了,这就需要做到趁早刷题。
提到刷题(怎么说这么多,逃),又会经常出现一种现象,也就是习得性无助。那种刷题刷到自闭怀疑人生,最终失去信心开始摆烂,这里我强 ...
欲戴王冠,必承其重
dc9fcc9c819d5a4f43fe8e7daa086cf61d64277f53779768f95dc77b0be8b39c105c36c28322a40f87bfe34ec821ccecc219cb76a9025884217b0a0b8e279771b769e25a971178ef2a5d16899edf242680612c2755385a04067006ecee72b0fa0fff1fd9ba04063a2c9c1978181574b2c6bcaf664b9d85736ee7ecf85607fa874deaed5179630046de71072342c4779fb25c0e609f7f2baabd82d484424e9b8fa90336d862330641b27870f7ab24059ba4cfdebc408d7e6e39ff1f785a38fbd6b215190e5c4d58cbf1e0e81e044d342f94711ed83556fe9581a595ca1864ec2ef89ef50097efd4dd673f06d81759898da92e724eced6a6330 ...
近况
dc9fcc9c819d5a4f43fe8e7daa086cf61d64277f53779768f95dc77b0be8b39c105c36c28322a40f87bfe34ec821ccecc219cb76a9025884217b0a0b8e279771b769e25a971178ef2a5d16899edf242680612c2755385a04067006ecee72b0fa0fff1fd9ba04063a2c9c1978181574b2c6bcaf664b9d85736ee7ecf85607fa874deaed5179630046de71072342c4779fb25c0e609f7f2baabd82d484424e9b8fa90336d862330641b27870f7ab24059ba4cfdebc408d7e6e39ff1f785a38fbd6b215190e5c4d58cbf1e0e81e044d342f8abce8fe9d326ad6ffe62f932c29f6ea79585bb1b7f55ce32b0f734fb4526faedea1c65b651895d9d ...
MIT ML
6.036 Machine learning
We are destined to meet again.
Overview or Big picture?
/loading
Introduction
This note will focus on the topics that mit note doesn’t cover so much, and some great ideas during the implement of HW.
Chapter 4 : Feature representation
The number of features becomes a problem if there are so many of them that some data points could have been identified by just looking at very few of their coordinates. Here it is not the case.
这部分的HW真的很棒,不同data形式对于最后的accuracy的影响的尝试与分析。
首先是 ...
Some talk about stochastic process
Background
When observing the natural world, many of us notice a somewhat beautiful dichotomy. No two things are ever exactly alike, but they all seem to follow some underlying form. Plato believed that the true forms of the universe were hidden from us. Through observation of the natural world, we could merely acquire approximate knowledge of them. They were hidden blueprints. The pure forms were only accessible through abstract reasoning of philosophy and mathematics. For example, the circle ...
Machine Learning Code
Just record the codes implement without using the library functions.
One algorithm One day.
Courses:
MIT_6.036
Python
Stanford_cs231n
这部分实在是太多了,后来直接就搬到github上了,想看完整版,可以去下面的link:
https://github.com/XuRui314/MIT_6.036_homework_zxr
1234567MIT's CAT~ML / /\__/\ mio~ \__=( o_O )= (____ ...
zxr青春的补完---Attack on Titan
本文为转载+自己思考
原视频为:全网最全面 进击的巨人解读!森林里的彷徨者!毁灭下的众生相【封神与烂尾】(下篇)_哔哩哔哩_bilibili
原作者:Lilin2白
zxr是从2013-2014看进巨的,也就是第一季动画化的时候,和表哥一起,基本从第一集开始追跟,阿尼巨人化的那一幕至今还带给我无比的震撼冲击,因为第二季和第一季差了好几年,所以就一直弃坑了7年左右吧,是大一的时候把漫画和动漫追到138话的时期,怎么说呢,前3季近乎完美,马莱篇似乎略显突兀和割裂,当时我是有点抗拒的,但是后来渐渐看下去,也似乎合理了一些,然鹅到后面剧情失控又有点迷惑了,最终jsc整了个离谱结局,加上b乎节奏暴乱,无意识地被带了节奏,就很唾弃这部作品,觉得是反战败和历史虚无主义的日本典型,基本认为是右翼作品。
然后就是2022刚刚的part2放送,一开始我是坚定不会去看,但是b站首页很离谱的给我推了很多part2的op、ed,加之身边朋友反映很棒,就只去看了ed和op,结果真的很惊艳,以艾伦的视角,那种释然和解,以及隐隐约约那份最初的感动,让我渐渐反思自己,消去了部分不理智,开始重新审视这部作品,不去神化 ...
Asuka
好多原图都是10M的那种,blog上都是压缩过的,可能不是很清晰。
但是找图这块zxr是不会让你们失望哒😋
画面光影色彩
虽然脸有点崩,但是色彩很喜欢
非常有感觉的画面和光影
很炫的香香
人物
夹带私货
涩图这块zxr也不会让你们失望哒)黄色连衣裙yyds
造型设计
说实话,联动和策划私货这块米忽悠是没让我失望过的。
场景
干杯[]–( ̄▽ ̄)*!
Self-criticism
dc9fcc9c819d5a4f43fe8e7daa086cf61d64277f53779768f95dc77b0be8b39c105c36c28322a40f87bfe34ec821ccecc219cb76a9025884217b0a0b8e279771b769e25a971178ef2a5d16899edf242680612c2755385a04067006ecee72b0fa0fff1fd9ba04063a2c9c1978181574b2c6bcaf664b9d85736ee7ecf85607fa874deaed5179630046de71072342c4779fb25c0e609f7f2baabd82d484424e9b8fa90336d862330641b27870f7ab24059ba4cfdebc408d7e6e39ff1f785a38fbd6b215190e5c4d58cbf1e0e81e044d342fe6ebfd29acfa2e6b124d7a54700a68150ce2e39aac5632f25386a1e794dd99caa00327d2f63153ba9 ...
java learning
Java Basic
learning web: https://ocw.mit.edu/ans7870/6/6.005/s16/index.html
Introduction to Java
While there are many IDEs with built-in execution capabilities (Eclipse and IntelliJ for example), Java can also be run directly from the command line.
Before you can run the program from the command line, you must compile it. Open your terminal or command prompt (depending on OS), and navigate to the directory where the file you want to run is located. Once there, use javac and the filename to com ...
English learning
Winter Holiday English Learning
Day 1
ted-learning-Do we see the reality as it is?
Summary
Do we see the reality as it is?
False Assumption: Perceptions are like reality, reality is like perception.
Mystery: The relation between brain activity and conscious experience.
It’s been a long time since the first batch of people thought about this problem. As Galileo’s interesting saying, do we just misinterpret the nature of our perceptions? To answer this question, from the perspective of modern sci ...
Artificial intelligence
Hi! This is zxr’s note of 6.034-artificial intelligence.😊
In general, it’s the record of lecture notes and zxr thinkings.
So, usually, i write the structure of lessons and inspiring ideas on paper when i watch the course. And later, i summarize them and generate the notes you see.
course and mega link:
6.034-course and 6.034-maga
zxr write most of the notes on paper, so this article is just like the supplement. Maybe zxr will scan and upload the paper note later. And i do recommend to read t ...