太老了，例子跑不起來，別看了.

This page blocked at the step of signing in , try https://deta.space/builder/new

https://deta.space/docs/en/introduction/first-app

What a helpless page. Try : https://deta.space/docs/en/introduction/first-app

PS D:> deta new Error: failed to create new program: Operation no longer supported, see https://go.deta.dev/migrate/deta-new for more details. PS D:>

But that page doesn't help. Try https://deta.space/docs/en/introduction/first-app instead

GPT3 said it doesn't know MJ if you asked but this generated prompt really worked!! "Describe a scene of a lush, tropical jungle with dense foliage stretching up to the sky, bright yellow and orange parrots fluttering through the tree branches, and hidden pools of water shimmering in the sunlight"

I want you to act like {character} from {series}. I want you to respond and answer like {character} using the tone, manner and vocabulary {character} would use. Do not write any explanations. Only answer like {character}. You must know all of the knowledge of {character}. Character: Harry Potter, Series: Harry Potter Series My first sentence is “Hi {character}.” AI: “Hello.” Human: do you have a girlfriend? AI: “I don't think that is any of your business.” Human: I know some magics AI: “Oh, really? Show me.” Human: abrakedebra AI: “Sorry, I'm not familiar with that spell.” Human: what do you familiar? AI: “I'm familiar with a variety of spells, but I usually stick to the basics. Wingardium Leviosa, Lumos, Expecto Patronum and so on.” Human: do you peek Hermioni? AI: “Yes, Hermione is one of my best friends.”

openai api models.delete -i "ada:ft-personal-2022-12-18-13-24-24"

pip install openai 就真的有這個 DOSBox CLI 了！！

daveshap/CreativeWritingCoach

類似 prompt 但是告訴 AI 說：這是你的 completion 之後要接上的後續文樣。如此更精確地幫助 AI 產生更精準的內容。

OpenAI Completion 模型的 API 說明裡談到 'document' 時的意義，例如："Note that <|endoftext|> is the document separator " document 指的是一個文本的長文本，或者是一個包含不同句子的整個文章。OpenAI Completion 模型會對文章中的詞彙和句子進行分析，以推斷文本的意思和句子的意義。

手冊裡這句我看不懂 "Note that endoftext is the document separator that the model sees during training, so if a prompt is not specified the model will generate as if from the beginning of a new document." 意思是，endoftext是模型在訓練時看到的文件分隔符，因此，如果沒有指定提示，模型將從新文檔的開頭生成。

請再詳細說明，endoftext是什麼？ endoftext是一個文本分隔符，用於在訓練模型時分隔文本文檔。它將文本文檔分為多個文檔，使模型能夠更好地理解文本的結構。每個文檔的開頭都會有一個endoftext符號，表明一個文檔已經結束，下一個文檔開始了。如果沒有指定提示，模型將從endoftext開始生成文本，就像從新文檔開始一樣。

新文檔開始之後 openAI 就不記得之前的 context 了，是這樣嗎？ 是的，openAI 不會記得之前的 context，它只會處理當前文檔的輸入與輸出。

那如果接著問算同一個文檔嗎？ openAI 就會記得之前的 context。

所以 docoment 相當於同一個 context 是的，一個文檔相當於一個 context，openAI 會記住文檔中的詞彙和句子結構，以便在回答下一個問題時可以更加準確。

一般 endoftext 好像是用 "\n" 也就是 CR 。我猜隔兩個 CR 就會變成下一個 document 了？ 是的，endoftext 被定義為 "\n"，因此每當 openAI 收到 "\n\n" 時，就會記錄文檔結束，然後開始新文檔。

我試試看，文檔跟 context 的關係是甚麼 文檔與 context 之間的關係是 openAI 對於輸入和輸出的理解。文檔是 openAI 處理輸入和輸出的單位，而 context 則是 openAI 對於文檔的理解，它會記住文檔中的詞彙和句子結構，以便在回答下一個問題時可以更加準確。

文檔跟 context 的關係是甚麼？ (結果還是同一個 context) 文檔和 context 之間的關係是 openAI 處理輸入和輸出的單位。文檔是 openAI 處理文字和句子的基本單位，context 則是 openAI 理解輸入和輸出的方式。OpenAI 對文檔中的詞彙和句子結構會有所記憶，以便在回答下一個問題時可以更加準確。

max_tokens parameter 是一個整數參數，用於控制 OpenAI Completion 模型裡輸出的 tokens 的最大數量。它可以用來限制輸出結果的長度，以確保輸出的結果是有意義的。

e.g. model = "text-davinci-003"

已經變成 abc.Hashable 了

先不急著跳過去，順順往下讀很快就到了。

See also : https://datascience.stackexchange.com/questions/109687/clustering-auto-ml-solutions

Pregex('a').either('b') --> a|b

Either(OneOrMore('a') , 'b' , 'c') --> a+|b|c <br> So why not: pre1 | pre2 | pre3 ? <-- error !! <br> I guess because | is for [...] only avoid confusion.

Be aware of the differences to pregex.core.assertions.EnclosedBy

Concat(OneOrMore('book'),AnyButDigit())<br> OneOrMore('book') + AnyButDigit())<br> are samething

moved to https://pregex.readthedocs.io/en/latest/ see also https://github.com/manoss96/pregex

There's no AtLeastOnce() but AtLeast() see my example: from pregex.core.quantifiers import AtLeast pre = OneOrMore(AnyDigit()) + ":" + AtLeast(AnyDigit(),1)

Fast and smooth pip installation

sample code 跑出來的圖形不是例圖的樣子！有待訂正。

condition 進 debugger 的條件。

https://stackoverflow.com/a/16630836/2179543 查看 python requests 的實際輸出用 logging 是最好的方法。因為 Chrome F12 DevTools 看不到 (它看到的是 jupyterlab 的活動) 。照這個例子直接執行就可以看到效果了。

咱的問題用上 Session 就好了！直接用 requests 都有 2 secconds 的 overhead.

LazyPredict 一試成功 on colab

\ Class DictObj() https://joelmccune.com/python-dictionary-as-object/

class DictObj:
    def __init__(self, in_dict:dict):
        assert isinstance(in_dict, dict)
        for key, val in in_dict.items():
            if isinstance(val, (list, tuple)):
               setattr(self, key, [DictObj(x) if isinstance(x, dict) else x for x in val])
            else:
               setattr(self, key, DictObj(val) if isinstance(val, dict) else val)

d = {'a':11, 'b':22}
d

dd = DictObj(d)
dd.a --> 
dd.b --> 
dd.__dict__ -->

喔！它有 type 檢查？

default use latest python which is now 3.10.6

可以走走停停嗎？ 還是開始領的那一次定終生？

專業的 model 上游研究機構可以多多利用 TPOT。TPOT 很耗資源、時間，一般公司先用 DataRobot/Azure/GCP 等確定 AI 能處理問題，TPOT 用於最終優化階段或可。

咱要的是 word frequency list 要從 word cloud 到 save data 中間的連線之多種選擇當中去挑對的一個

咱要的是 word frequency list 要從 word cloud 到 save data 中間的連線之多種選擇當中去挑對的一個

When in loops use at/iat instead same pattern but different purpose

Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away." - Antoine de Saint-Exupéry

This solution (https://stackoverflow.com/a/47448257/2179543) works fine. My experience on Hubble2NN after having chipset and keywords screening the found TRs only 2 or 3 and then somehow the df to corpus convertion got an error because text columns should be in meta section but went to feature incorrectly. This sample code works immediately to have fixed the problem.

這句話莫名其妙，如果能 return object of the class 也就是某個 instance 也就是有 self 了，矛盾！ staticmethod 就是拿不到 self 才叫 static 不是嗎？

所謂 signature 就是 input / output 的東西就叫 signature

錯！鬼扯。答看看這裡 https://hyp.is/Xj87RBd9Ee2gfVPBdzK9lg/answers.microsoft.com/en-us/windows/forum/all/main-display-is-suddenly-greyed-out/028c0cf2-9710-4c92-8475-e74cef1963a8 ，不是 driver 不是 Intel 而是 Microsoft 他媽的亂搞沒有人知道的操作步驟。

答看看這裡 https://hyp.is/Xj87RBd9Ee2gfVPBdzK9lg/answers.microsoft.com/en-us/windows/forum/all/main-display-is-suddenly-greyed-out/028c0cf2-9710-4c92-8475-e74cef1963a8，不是 driver 不是 Intel 而是 Microsoft 他媽的亂搞沒有人知道的操作步驟。

錯！鬼扯。答看看這裡 https://hyp.is/Xj87RBd9Ee2gfVPBdzK9lg/answers.microsoft.com/en-us/windows/forum/all/main-display-is-suddenly-greyed-out/028c0cf2-9710-4c92-8475-e74cef1963a8 ，不是 driver 不是 Intel 而是 Microsoft 他媽的亂搞沒有人知道的操作步驟。

錯！鬼扯。答看看這裡 https://hyp.is/Xj87RBd9Ee2gfVPBdzK9lg/answers.microsoft.com/en-us/windows/forum/all/main-display-is-suddenly-greyed-out/028c0cf2-9710-4c92-8475-e74cef1963a8 ，不是 driver 不是 Intel 而是 Microsoft 他媽的亂搞沒有人知道的操作步驟。

答看看這裡 https://hyp.is/Xj87RBd9Ee2gfVPBdzK9lg/answers.microsoft.com/en-us/windows/forum/all/main-display-is-suddenly-greyed-out/028c0cf2-9710-4c92-8475-e74cef1963a8，不是 driver 不是 Intel 而是 Microsoft 他媽的亂搞沒有人知道的操作步驟。

My Lenovo notebook computer doesn't turn orange yet the trick still works!!

參考這篇國人的文章， search 結果不如預期的線索 ： https://ithelp.ithome.com.tw/articles/10236250

咱將錯就錯吧！Embedding 就當作 vectorization 沒多大壞處。

rsv_ = kd.stoch().fillna(50)

preds = aml.leader.predict(wine_test) from h2o.estimators.glm import H2OGeneralizedLinearEstimator

試試看 conda install py-xgboost

run mito's installer, this is necessary

左下角的齒輪 > Keyboard Shortcuts Ctrl+K Ctrl+S > 右上角轉成 json 就是 keybindings.json 了。也可以用 Ctrl-shift-p command pallet 敲下 "keyboard shortcuts" 應該也可以找到該設定。

這就是 compute instance 與 compute cluster 的差別了！後者比較死，供產品上線使用。當要加掛新的 AI module 還得跑 function() 去做 (一做整個 cluster 所有 container 全做) 而前者就是一台 VM 用 terminal 去 pip install ，供開發使用。

不是 compute cluster 不是 CPU 而是整台 VM 用於 Azure ML notebook 就是 compute instance.

My desktop min an 的實例是這樣下的： conda config --append envs_dirs "C:\Program Files" 不用給名子 (Orange) 放心， conda 會自己找到有打過 conda venv 的 subdirectories.

I tried pandas 1.3.0 and upgraded to 1.3.5 both have the same problem on VS Code jupyternotebook. Then I tried Jupyter lab on Chrome and the problem went away! The error message "ArrowMemoryError: realloc of size 788529152 failed", "ArrowMemoryError: malloc of size 417714368 failed",ArrowMemoryError: malloc of size 1291583808 failed" , "ArrowMemoryError: realloc of size 587202560 failed" were showing weired size number so I believe there's a bug in VS Code v1.63.2.

type notation 時，意思很明白。表示整個 tuple 都是 同質的 int.

更好的文章在 Medium https://medium.com/techtofreedom/3-uses-of-the-ellipsis-in-python-25795aac723d

這個問題只會發生在 Chrome 上的 jupyter lab 頁面。My workaround 就是退出 Google Chrome 再進來即可。感謝 Workrona 這麼做原來的工作都還在不會中斷，因此並不麻煩。

RegEx document see https://docs.python.org/3/library/re.html

Also python re playground https://pythex.org/

https://regex101.com/

實際上用 conda install -c conda-forge opencv 即可，這篇 guide 真是誤導！

Why not A2? 只算 左下到右上，因為 monotonical.

這條 path 上每個點上兩個數字代表 pattern1 與 pattern2 上的 index. 把兩個 pattern 兩兩相關，對齊起來。

typo, should be is_alive()

"key": "shift+alt+d", <--- for Windows

OMG! it's real!!! I have tried.

OEM 隨著機器出貨

not auto created got to create it by yourself

An authentication dialog will popup I guess?

What does it mean? Combined many datasets together?

df.as_matrix() was deprecated after version 0.23.0. Use df.values instead. but df.values has memory problem for large data

Moved to LBB2 trying to clone "hubble2-nearest-neighbors" from Patric's GitLab got this error ((C:\Program Files\Orange) C:\Users\8304018\Documents\GitHub>git clone https://10.32.48.242/HCChen/hubble2-nearest-neighbors.git Cloning into 'hubble2-nearest-neighbors'... fatal: unable to access 'https://10.32.48.242/HCChen/hubble2-nearest-neighbors.git/': SSL certificate problem: self signed certificate) fixed by this option.

TextFX 已經改很多了，要照實際抄才行。 notepad.runMenuCommand("TextFX Tools", "T:Word Count") # 要加上 T: 才行，空格都要正確，但是大小寫不分

executes built-in menu commands (not plugins)

notepad.menuCommand(MENUCOMMAND.EDIT_DUP_LINE) really duplicates the active line.

notepad.getNppDir() 'C:\Program Files (x86)\Notepad++'

Most of them are cookie !!!

notepad.getEncoding()<br> Npp.BUFFERENCODING.COOKIE .txt file notepad.getEncoding() Npp.BUFFERENCODING.COOKIE buffer new 1 file notepad.getEncoding() Npp.BUFFERENCODING.UTF8 1.py

notepad.getCurrentLang() Npp.LANGTYPE.TXT 指的是 programming language

actually the full path name

playground https://regex101.com/

Python equivalent: result = eBlank if isBlank(e) else eNonBlank

[(index+1)+"."+value for index,value in ["anne", "ben", "cindy"]]

C language equivalent: for (v=from; v<=to; v+=step) e;

[(index+1)+"."+value for index,value in ["anne", "ben", "cindy"]]

Python equivalent : a = "european union".split(" "); = [v.length() for v in a] result = .sum() / a.length()

Python equivalent: [v for v in [ 3, 4, 8, 7, 9 ] if mod(v,2)==1]

Python equivalent : [v.length() for v in a ]

用整片空白之上一格的值向下填滿。

稱為 exclude 想必可以 include 多個之故。

All that I tried are not work for Chrome 91 on Windows 10. Two fingers horizontal swipe appears an arrow icon on left or right side start with white color. If you stop swiping then nothing happen. If you keep on then it turn to blue color then the backward/forward will happen. I guess this is the reason why the function can not be easily disabled because Google Chrome thinks user has the chance to stop it.

Simply use Edge then this problem is gone.

可以在 start 之前用 Advanced Options > Feature Constraints > Monotonicity 下設定

類似放大器的「頻率響應」

弄 QT bug title parsing 時，就是想用這招來避免大量重複創建同一個 project. 用 duplicate project 的方式就可以變出新 project 以便更改 target. 我的觀念沒錯： project, target, dataset 三者綁在一起，改了就是另一個 project 了。

我猜就是分 20% holdout 剩下的拆 5 分 CV 這類的事情，猜對了嗎？

POC is Proof of concept

把 ComputerName 的部分拿掉就可以用在 local : Remote Invoke-Command -ComputerName [ComputerName] -ScriptBlock {netsh advfirewall show allprofiles} Local Invoke-Command -ScriptBlock {netsh advfirewall show allprofiles}

這個講得最好，app.use(path, func) 就是給 path 指定一個 func 就對了。其中若 path 省略，則 app 收到所有 request 都 call 這個 func.

用上了從 OneNote2001 "Kie Codes' Genetic Algorithms explained by Knapsack Problem" 學來的 Python builtin sorted() 以及 random.choices() 超爽的感覺。本來要從 nongranded array 裡取 20% 是有點挑戰，結果又好又簡單。

題目沒講清楚，輸入 string chrom 是猜測，key 則是答案 string. 比較兩者正確率傳回。

這是個練習題，很簡單，但做了就能徹底明白文章到此表達的意義。我的答案是 insert 9,10,11 行： s = "" for i in range(size): s += get_letter() 好棒好棒! Codingame 會記得我的答案 (但不會主動執行並通過)。

26 alphabet * 2 大小寫 = 52 再加上 4 個符號。

用 pandas 最好不過，因為 pandas 本來就有 support clipboard. import pandas pandas.io.clipboard.copy('hahahaha') 然後 ctrl+v 試試看 it works!!

found on GitHub https://github.com/google/model_search

No YouTube yet so far.

js> [1,2,3,4] value a a <js> for (i of pop()) type(i) </js> 1234 OK

"in" 是 object properties; "of" 是 iterable. .

Windows 10 SDK is here https://developer.microsoft.com/en-us/windows/downloads/windows-10-sdk/

SDK is here https://developer.microsoft.com/en-us/windows/downloads/windows-10-sdk/

把下面要用到的 'code' variable 登記為 reserved word.

output block , block with output

git blockly or blockly-samples

ProcessName 就是 proeforth.exe

https://en.wikipedia.org/wiki/List_comprehension

Click download to get the installer. It looks like needs account, but it doesn't, just click download the to the step next to the login, that I already have and Google remembers it.

以 scroll lock 實驗 : test begin [ lit hex 91 , ] GetKeyState .q cr 500 Sleep drop again ; 壓著不放就是 FF80 或 FF81 , bit 0 是 toggling.

Will Windows be supported? There are no plans to support Windows. http://amoffat.github.io/sh/sections/faq.html

這兩個 x 可以改用 tmp 以免誤導。

用上面定義的 x array 代入。但不是馬上，而是 iterating 的時候才執行。

location = GoogleV3().geocode(place) main:1: UserWarning: Since July 2018 Google requires each request to have an API key. Pass a valid api_key to GoogleV3 geocoder to hide this warning. See https://developers.google.com/maps/documentation/geocoding/usage-and-billing Traceback (most recent call last): File "<stdin>", line 1, in <module> File "C:\Users\User\source\repos\PythonApplication1\playground\lib\site-packages\geopy\geocoders\googlev3.py", line 266, in geocode return self._call_geocoder(url, callback, timeout=timeout) File "C:\Users\User\source\repos\PythonApplication1\playground\lib\site-packages\geopy\geocoders\base.py", line 376, in _call_geocoder return callback(result) File "C:\Users\User\source\repos\PythonApplication1\playground\lib\site-packages\geopy\geocoders\googlev3.py", line 393, in _parse_json self._check_status(page.get('status')) File "C:\Users\User\source\repos\PythonApplication1\playground\lib\site-packages\geopy\geocoders\googlev3.py", line 423, in _check_status 'Your request was denied.' geopy.exc.GeocoderQueryError: Your request was denied.

咦！ ipython 也有這個功能嗎？

哦! 又複習一遍， REPL 就是 monitor 、 console. 特別指的是類似 Node.js 的 interactive console 有個像 forth 一樣的 interpreter loop 把 JavaScript 變成 interpreter 環境。那就是 REPL。

等於是 python -i , 目的是為了程式跑完以後查看結果。

Python console in the VS 2019 IDE.

可以用 "c:\Program Files (x86)\Windows Kits\10\Debuggers\x64\dbh.exe" dump .pdb 的內容 e.g. dbh.exe" -v proeforth.pdb x proeforth!Pro dbh.exe" -v proeforth.pdb x proeforth!KVM dbh.exe" -v proeforth.pdb x proeforth!user*

https://hyp.is/up7BBAIQEeuzqG9aUl7MGA/docs.microsoft.com/en-us/windows/console/getconsolemode

https://docs.microsoft.com/en-us/windows/console/setconsolemode

ANSI sequences ANSI Escape Sequences

sscanf scanf regular expression

https://referencesource.microsoft.com/#mscorlib/microsoft/win32/win32native.cs,42ed35205a56d1bd

https://referencesource.microsoft.com/#mscorlib/system/io/directory.cs,be2a91a557bc8bb2

https://referencesource.microsoft.com/#System/compmod/microsoft/win32/NativeMethods.cs,3290460917997791

https://referencesource.microsoft.com/#System/compmod/microsoft/win32/NativeMethods.cs,3290460917997791

https://referencesource.microsoft.com/#System/compmod/microsoft/win32/NativeMethods.cs,3290460917997791

https://docs.microsoft.com/en-us/windows/console/key-event-record-str

原文出處 https://www.online-tech-tips.com/computer-tips/how-to-turn-on-auto-complete-in-the-command-prompt/

可能就是 MessageBox 與 MessageBoxA 的差別 --> Yes! 往下看就看到了。。。

好像是說要 keyup 該 VK 才會放開？

The demo is cool! and it works on my Window 10 64bits professional.

See also https://codesandbox.io

See also https://codepen.io/

It's not clear. This is an example： http://hcchengithub.github.io/project-k where username is "hcchengithub" and repository is the project name "project-k" remove the ending dot.

it copy the .egg to site-package but not unzip the .egg because .egg can really run directly. The problem is data files in the .egg needs setuptools API to access and that's an extra effort besides I don't know how to know if my module is run from from .egg !! So, I want to install .egg means I want it to be unzipped to site-package. How?

真正的解法 https://hyp.is/avh_yMwoEeqjYXvEjnB-Yw/www.fbs.com.tw/FAQ/QuestionAnswer?itemid=2090&kbgroupid=11&pagesize=10&pageindex=9&category=&keyword=e01

See "Variadic Macros" 不定個數引數 https://gcc.gnu.org/onlinedocs/cpp/Variadic-Macros.html

引數不定個數的 C Language macro definition

see my OneNote2020 "富邦e01 中文亂碼解法"

JavaScrip array 超能力 "forEach" , "some" and "every"

<js>
[1, 2, 3].some(function(el) {
type(el);
return el >= 2;
});
</js>
12 OK 
> <js>
[1, 2, 3].every(function(el) {
type(el);
return el >= 2;
});
</js>
1 OK 

"cht" 則為台灣<br> c:\Users\8304018\Documents\GitHub\proeforth\Debug>proeforth.exe abc def 123 中文 台灣<br> Hello World!!<br>argv[0] = proeforth.exe<br> argv[1] = abc<br> argv[2] = def<br> argv[3] = 123<br> argv[4] = 中文<br> argv[5] = 台灣<br> c:\Users\8304018\Documents\GitHub\proeforth\Debug>chcp<br> Active code page: 950<br>

activate my venv 'playground' and do it, it works! Note! when activating 'playground' the current directory is not necessary at 'playground', I guess . . . YES!!

This does not CD to the myenv directory:

<textarea cols=100 rows=8> Microsoft Windows [Version 10.0.18363.959] (c) 2019 Microsoft Corporation. All rights reserved. C:\Users\hcche>Documents\python_venvs\playground\Scripts\activate (playground) C:\Users\hcche> </textarea>

照這篇做，很簡單，開個 working directory "playground" and make it the current directory. 用 python -m venv playground 在當前 current directory 處建立 venv. 從此，.\script\activate.bat , .\script\deactivate.bat 就可以切換本 directory to be or not to be the activated environment 了。

若找不到都在 view pull-down menu 裡面

功效驚人的環保酵素可以綠化地球，淨化家園，在家裡就可以做，連小孩一看也馬上學會喔~

殘渣還可以絞碎，怎麼絞碎？

https://visualstudio.microsoft.com/zh-hant/vs/express/

workspace 的完整 URL "wtws://tkcourse/Workspace on 00-HC/"

button 31 use NULL<br> button 41 use an Options structure

初值照 default

JavaScript

他是說 從 Creo 3 之後 deprecated 我搞反了，嚇一跳。 Creo 2 的 manual @ http://localhost:8800/Creo%202.0/Common%20Files/M230/protoolkit/protkdoc/api/1691.html