Welcome all! In the first episode of this series, I investigated the four most known machine learning frameworks and discussed which of these you should learn depending on your needs and goals.

w ^迎閱讀所有！ 在本系列的第一集中 ，我研究了四種最著名的機器學習框架，并討論了根據您的需求和目標應該學習的機器學習框架。

Of course, we all know how PyTorch and TensorFlow are overwhelmingly effective and thorough when it comes to building deep learning algorithms from scratch. Similarly, Scikit-learn comes with excellent non-neural solutions and a whole lot of convenient data processing and evaluation functions.

當然，當我們從頭開始構建深度學習算法時，我們都知道PyTorch和TensorFlow如何具有壓倒性的優勢和全面性。 同樣， Scikit-learn具有出色的非神經解決方案以及大量便捷的數據處理和評估功能。

Today, I would like to draw your attention to something different: a rising machine learning framework, Libra, that absolutely deserves your attention and has tripled its amount of GitHub stars in only a week (reaching 1.8K)!

今天，我想提請您注意一些不同的東西：新興的機器學習框架Libra絕對值得您的注意，并且在短短一周內將GitHub star的數量增加了兩倍(達到1.8K)！

In this article, I will cover:

在本文中，我將介紹：

1. The basic concept of the framework

   框架的基本概念

2. The interesting features I spotted

   我發現的有趣功能

3. The expected downsides

   預期的缺點

4. A concrete use example

   具體的使用示例

一，基本概念 (I. Base Concept)

Libra is an easy-to-use machine learning framework that will allow you to load data, process it, train models and visualize results with only a few lines of code.

Libra是一個易于使用的機器學習框架，使您僅需幾行代碼即可加載數據，處理數據，訓練模型并可視化結果。

The major advantage of this framework is its user-friendliness and beginner-friendliness. The diagram below shows about how fast you can setup an ML project compared to more common frameworks:

這個框架的主要優點是它的用戶友好性和初學者友好性 。 下圖顯示了與更常見的框架相比，您可以更快地設置ML項目：

This literally means that you can come with no or nearly no technical machine learning knowledge and implement a full project in about 5 minutes.

從字面上看，這意味著您可能沒有或幾乎沒有任何機器學習知識，而僅需5分鐘即可實現一個完整的項目。

The library is built for Python (?3.6) and revolves around a Client object that will handle your data but also the models you want to build, inference, plotting, etc.

該庫是為Python(?3.6)構建的，圍繞一個Client對象運行，該對象不僅可以處理數據，還可以處理要構建，推斷，繪制等的模型。

You can easily install and initialize the library in a project with only lines of code: first, you need to install Libra and all the dependencies in your environment with pip install libra. Next, simply write from libra import client in a Python file and you’re all set!

您只需使用幾行代碼即可輕松地在項目中安裝和初始化該庫：首先，您需要使用pip install libra安裝Libra以及環境中的所有依賴項。 接下來，只需from libra import client編寫一個Python文件即可，一切就緒！

二。 有趣的功能 (II. Interesting Features)

The first interesting feature I noticed is that some “queries” try to infer what is asked of them: for example, if you call a neural_network_query(), the only required argument is a quick explanation of what you want to do, like “please model the median number of households”.

我注意到的第一個有趣的功能是一些“查詢” 嘗試推斷出他們的要求 ：例如，如果您調用Neuro_network_query() ，則唯一需要的參數是對您想做什么的快速說明，例如“請模擬住戶中位數”。

The target variable in your data will then be determined through the parsing of your explanation and computing the levenshtein distance on the column names.

然后，將通過解析您的解釋并計算列名上的levenshtein區別來確定數據中的目標變量。

Kind of cool, right?

有點酷吧？

Moreover, you can simply load your .csv data file and it will automatically be handled in a Pandas DataFrame and preprocessed for you during the query, which is convenient if you want fast results!

此外，您只需加載.csv數據文件，它將在Pandas DataFrame中自動處理并在查詢過程中為您進行預處理，如果您想要快速的結果，這將非常方便！

The fact that Libra relies on other machine learning libraries means that all ML tasks are virtually doable. Here is a list of the currently available queries, and includes Feedforward NN, CNN, SVM, Text generation, Sentiment analysis, etc.

Libra依賴于其他機器學習庫這一事實意味著所有ML任務實際上都是可行的。 這是當前可用查詢的列表，包括前饋NN，CNN，SVM，文本生成，情感分析等。

When you fine tune a model in with the Libra client, the evaluation results and plots are automatically displayed for you. Here is an example where I tried to predict sentiment scores of tweets from this Kaggle dataset using a simple neural network:

當您使用Libra客戶端微調模型時，評估結果和圖將自動為您顯示。 這是一個示例，我嘗試使用簡單的神經網絡從該Kaggle數據集中預測推文的情感評分：

The project is growing pretty fast and even though the core development has been done, you can put your skills to use and contribute by adding missing bits of code and correcting bugs!

該項目正在Swift發展，即使完成了核心開發，您也可以通過添加缺少的代碼并更正錯誤來利用和貢獻自己的技能！

三， 缺點 (III. Downsides)

The major downside you will experience with an ML framework that is so beginner-friendly is process over-simplification: by providing such high-level wrapping techniques, you sort of lose the sense of control over what is happening.

ML框架對初學者非常友好，這將給您帶來的主要缺點是過分簡化了流程 ：通過提供這種高級包裝技術，您會失去對正在發生的事情的控制感。

For example, when using the “neural network query”, the number of distinct target variables in your data will determine if a classification or regression task will be performed, and you will not have a choice in the neural net’s architecture details.

例如，當使用“神經網絡查詢”時，數據中不同目標變量的數量將確定是否要執行分類或回歸任務，并且您將無法選擇神經網絡的體系結構詳細信息。

As of today (August 2020), Libra is still being developed by the creators and the community, meaning that a lot of bugs appear here and there, which can be discouraging at first.

截至今天(2020年8月)，天秤座仍由創作者和社區開發，這意味著到處都會出現許多錯誤 ，一開始可能會令人沮喪。

Even though Libra is a relatively new machine learning framework, it hasn’t (yet) re-invented the wheel, and relies on popular libraries like Keras, Transformers, Scikit-learn and NLTK. This mainly means that the number of hard dependencies is quite high, and can be an issue if you like to keep a light environment.

盡管Libra是一個相對較新的機器學習框架，但它(尚未)重新發明輪子，而是依賴于Keras，Transformers，Scikit-learn和NLTK等流行的庫。 這主要意味著硬依賴性的數量非常高 ，如果您希望保持一個輕量級的環境，則可能會成為一個問題。

IV。 簡單項目示例 (IV. Simple Project Example)

To show you how easy and fun it is to use Libra, let’s implement a small project together: based on the beginning of J.R.R Tolkien’s Silmarillion, we will generate new text with only 3 lines of code.

為了向您展示使用Libra是多么容易和有趣，讓我們一起實現一個小項目：基于JRR Tolkien的Silmarillion的開頭，我們將只用3行代碼生成新文本 。

from libra import clienttolkien_client = client('silma_beg.txt')
tolkien_client.generate_text("generate some text please", return_sequences=1)
print(tolkien_client.models['text_generation']['generated_text'])

Libra will load GPT-2 from OpenAI and generate new text for us.

天秤座將從OpenAI加載GPT-2并為我們生成新文本。

In the following Gist, the first line is actually the input file I fed the network with. All the rest was automatically output by the model, and it’s pretty amazing!

在下面的要點中，第一行實際上是我為網絡提供的輸入文件。 其余的全部由模型自動輸出，這真是太神奇了！

There was Eru, the One, who in Arda is called Ilúvatar; and he made first the Ainur, the Holy Ones, that were the offspring of his thought, and they were with him before aught else was made. And he spoke to them, propounding to them themes of music; and they sang before him, and he was glad. But for a long while they sang only each alone, or but few together, while the rest hearkened; for each comprehended only that part of me mind of Ilúvatar from which he came, and in the understanding of their brethren they grew but slowly. Yet ever as they listened they came to deeper understanding, and increased in unison and harmony. And the time came when they saw that, at last, by the love which the people gave her, there had been a way of getting over her by the love of her Creator. And they arose from their hiding place, and sought the land in the wilderness, and there they were put to the slaughter, for they were of the blood of the sons of Ilúvatar, and it was not easy to get over them. For those who had been left by their parents, as it were, had come to know the true nature of what they had done, and to believe that they had done what had been done, and that they had found the love of the Holy Ones; but Ilúvatar said to the other:Thy sons are born of the Father, and thou shalt have their hearts at my feet, and thou shalt have my prayers to thee in the heavens.And the angels said to Ilúvatar:Father, take these people, O Ilúvatar; make them come back.And Ilúvatar said unto them:Thou are worthy, O Ilúvatar, of these sons.And when they were come back to their hiding place, they were put to the slaughter; for they had seen that their father was not in the way of righteousness, but rather from a power of the Holy One, whom they saw the sons of Ilúvatar.And the angels, then, said unto Ilúvatar:The love of God, Ilúvatar, is strong and great; and thou shalt be glad, O Ilúvatar, if thou hast the Holy One with you. For there are few who ever have been so much as children, and

The model probably was pre-trained on the entire book, but the output grammar and generation relevance is nonetheless very good.

該模型可能已在整本書中進行了預訓練，但是輸出語法和生成相關性仍然非常好。

This is what Libra is about: creating with only a handful of lines of code. The opportunity to load, train, infer, evaluate and plot without prior extended knowledge.

這就是Libra的目的： 只用幾行代碼創建 。 在沒有事先擴展知識的情況下進行加載，訓練，推斷，評估和繪圖的機會。

I hope you have enjoyed this article, thank you very much for reading through and make sure to try out this framework when you get the chance!

我希望您喜歡這篇文章，非常感謝您通讀，并確保在有機會的情況下嘗試使用此框架！