快速數(shù)據(jù)庫框架

重點(diǎn) (Top highlight)

數(shù)據(jù)科學(xué) (Data Science)

Success in data science and software engineering depends on our ability to continuously learn new models and concepts.

數(shù)據(jù)科學(xué)和軟件工程的成功取決于我們不斷學(xué)習(xí)新模型和概念的能力。

Both domains are infinitely large umbrellas of nested ideas.

這兩個領(lǐng)域都是嵌套思想的無限大傘。

While you can spend decades in a specific branch of NLP, many of use are generalists and need to constantly acquire new knowledge.

盡管您可以在NLP的特定分支上花費(fèi)數(shù)十年的時間，但許多用途是通才，需要不斷獲取新知識。

Here’s my framework for doing this quickly.

這是我快速執(zhí)行此操作的框架。

不要以數(shù)學(xué)或科學(xué)論文開頭 (Do not start with math or scientific papers)

Math is overrated.

數(shù)學(xué)被高估了。

這對兩件事很有用： (It’s useful for 2 things:)

It gives an intuition for how things work.

它為事物的工作方式提供了直覺。

It allows coding algorithms without using an existing package

它允許編碼算法而無需使用現(xiàn)有程序包

軼事： (Anecdotally:)

The former is useful when tuning models.

前者在調(diào)整模型時很有用。

The later is useful when I need an algorithm in a Ruby but packages supporting it only exist in Python.

當(dāng)我需要Ruby中的算法但是支持它的程序包僅存在于Python中時，后者很有用。

Math is never the first step.

數(shù)學(xué)絕不是第一步。

觀看youtube視頻，以高水平掌握它 (Watch a youtube video to grasp it at a high level)

Watch a couple videos. You can find them covering most concepts.

觀看幾個視頻。 您會發(fā)現(xiàn)它們涵蓋了大多數(shù)概念。

Don’t try to understand specific technicalities at this point. Just develop a high level idea of what it does and how it’s different from related concepts.

此時不要嘗試了解特定的技術(shù)。 只要對它的功能以及與相關(guān)概念的不同之處有一個高級的了解即可。

Videos are great because images cut through the technical jargon that often makes written explanations seem more complicated than reality.

視頻之所以如此出色，是因為圖像突破了技術(shù)行話，往往使書面說明顯得比現(xiàn)實復(fù)雜。

Get to the point where you can describe the concept in a few sentences. At this point, it’s useful imagining that you’re explaining it to a non-technical colleague.

到達(dá)可以用幾句話描述這個概念的地步。 在這一點(diǎn)上，想象您正在向非技術(shù)同事解釋它很有用。

使一些代碼正常工作 (Get some code working)

Find a code snippet online.

在線查找代碼段。

You can often google “concept” + “python tutorial” to find code. Otherwise, find a related library and review the docs, or search related tags on Stack Overflow.

您通常可以通過Google“概念” +“ Python教程”來查找代碼。 否則，找到一個相關(guān)的庫并查看文檔，或在Stack Overflow上搜索相關(guān)的標(biāo)簽。

Expect to wrestle with conflicting packages and APIs that differ between versions at this point. Always use a virtual environment to keep your machine clean!

期望此時解決版本之間不同的沖突軟件包和API。 始終使用虛擬環(huán)境來保持機(jī)器清潔！

Get some code working. Change variables. Break it.

使一些代碼正常工作。 更改變量。 打破它。

Look at how the data changes at each step.

查看數(shù)據(jù)在每個步驟中如何變化。

You’re laying the groundwork that you’ll mentally attach a deeper understanding to later.

您正在奠定基礎(chǔ)，以后將在頭腦上加深了解。

分解概念中的步驟 (Break down the steps in the concept)

Go back to youtube.

返回youtube。

Write out its steps on paper. Draw a flowchart and revise it as you better understand the concept.

在紙上寫下它的步驟。 繪制流程圖并對其進(jìn)行修改，以使您更好地理解該概念。

High-level — what components and different steps occur?

高級別-發(fā)生了哪些組件和不同的步驟？

Isolate the setup (preprocessing) from the model itself that you’re learning.

從您正在學(xué)習(xí)的模型本身中隔離設(shè)置(預(yù)處理)。

(可選)復(fù)習(xí)數(shù)學(xué)并閱讀論文 ((Optional) Review the math and read papers)

Dig into each step.

深入每一步。

Understand the math.

了解數(shù)學(xué)。

This is most beneficial if you already have experience with adjacent concepts. Reviewing an advanced paper in a domain you have no experience with will sink A LOT of time.

如果您已經(jīng)有相鄰概念的經(jīng)驗，這將是最有益的。 在您沒有經(jīng)驗的領(lǐng)域中審閱高級論文會浪費(fèi)很多時間。

Now lay the math on top of your previous steps. Again, Khan Academy and YouTube can be helpful here.

現(xiàn)在，將數(shù)學(xué)放在您之前的步驟之上。 同樣，可汗學(xué)院和YouTube在這里可能會有所幫助。

我通常僅在以下情況下采取此步驟： (I typically only take this step if:)

An MVP (base-case) is complete and is ready to be optimized

MVP(基本情況)已完成，可以進(jìn)行優(yōu)化了

It’s genuinely interesting

真的很有趣

I need to evaluate tradeoffs between technical options

我需要評估技術(shù)選擇之間的權(quán)衡

用它建造東西 (Build something with it)

Use it or lose it.

使用它或失去它。

Personally, I don’t remember anything I only read about.

就個人而言，我不記得我只讀過的任何東西。

Apply it to your own use-case. The important piece here is to apply it in a completely different situation than the example you learned it on.

將其應(yīng)用于您自己的用例。 這里重要的一點(diǎn)是將它應(yīng)用于與您所學(xué)到的示例完全不同的情況。

While reviewing the landscape of concepts in a domain is useful, it’ll be more readily available to you (when you need it) if you’ve applied it previously.

回顧領(lǐng)域中概念的概況很有用，但是如果您之前已經(jīng)應(yīng)用過它，那么在需要時將更容易使用。

Build something simple.

構(gòu)建簡單的東西。

結(jié)論 (Conclusion)

That’s it.

而已。

Additionally, learn things adjacent to what you already know. Foundations are underrated. You can’t learn calculus without basic arithmetic.

此外，學(xué)習(xí)與您已經(jīng)知道的知識相鄰的事物。 基金會被低估了。 沒有基本的算術(shù)就無法學(xué)習(xí)微積分。

Now go learn some stuff. And use it (for good)!

現(xiàn)在去學(xué)習(xí)一些東西。 并使用它(永遠(yuǎn))！

翻譯自: https://towardsdatascience.com/a-framework-for-learning-new-data-science-concepts-quickly-4a691250dc5c

快速數(shù)據(jù)庫框架

總結(jié)

以上是生活随笔為你收集整理的快速数据库框架_快速学习新的数据科学概念的框架的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。