当前位置：网站首页>In 2022, you still can't "low code"?Data science can also play with Low-Code!

In 2022, you still can't "low code"?Data science can also play with Low-Code!

2022-07-29 21:39:00 【ShowMeAI】

作者：韩信子@ShowMeAI
数据分析实战系列：http://www.showmeai.tech/tutorials/40
机器学习实战系列：http://www.showmeai.tech/tutorials/41
本文地址：http://www.showmeai.tech/article-detail/286
声明：版权所有,转载请联系平台与作者并注明出处
收藏ShowMeAI查看更多精彩内容

低代码开发,顾名思义,Refers to the software development process only need to write a small amount of code is enough.与传统开发方式相比,Low-code drastically reduces the effort to write code,This makes it faster、更短的开发时间与更低的成本.

无代码 / Low-code machine learning platform（和库）的兴起,Speed up code development.With the help of these platforms and frameworks,Data scientists are between heavy exploratory research and heavy programming tasks,achieve a better balance.

在本篇内容中,ShowMeAI Summarize the most worthwhile learning for you&使用 Python 低代码机器学习库,Covers the most popular directions in data science——数据分析&simple mining、机器学习、深度学习.

数据分析&simple mining

D-Tale

D-Tale is an easy to use low code Python 库,通过将 Flask Backend written with React Combining with the front-end written,与 Jupyter Notebook 无缝集成,可以查看和分析 Pandas 形态的数据,包括 DataFrame、Series、MultiIndex、DatetimeIndex 和 RangeIndex.

D-Tale 是 SAS 到 Python product of conversion,最初是基于 SAS 的 perl 脚本包装器,现在是基于 Pandas Lightweight data structures Web 客户端.

大家可以在D-Tale的官方 Github Check out its detailed tutorial and usage,也可以前往在线平台操作体验.

AutoViz

For low-code exploratory data analysis tasks,AutoViz 是 Python Another good choice.在功能方面,It can be used with just one line of code AutoViz Complete automated visualization of any dataset.

AutoViz Ability to combine tasks to determine which features are most important,The information is then drawn and presented by using only those automatically selected elements.而且AutoViz速度极快,Visualization can be done in seconds.

You can check the official one AutoViz 示例 Jupyter Notebook 进行学习.

Lux

Lux ToolPak is a very automatic data analysis visualization tool.No need to do much data preprocessing,It automatically generates a series of candidate charts based on the data,You can choose from them according to actual needs.This greatly reduces the time required to make graphs and the amount of data preprocessing.

大家可以通过 Lux 的官方 GitHub page for more usage details.

Pandas-Profiling

pandas-profiling Libraries are automatically targeted pandas DataFrame Format data to generate data analysis reports.

The end result is interactive HTML 报告呈现,包含以下信息：

类型推断：字段列的类型
要点：类型、唯一值、缺失值
分位数统计：包括最小值、Q1、中位数、Q3、最大值、范围、四分位间距
描述性统计：包括均值、众数、标准差、总和、中值绝对差、变异系数、峰度、偏度等
直方图：分类和数字
相关性：Spearman、Pearson 和 Kendall 矩阵
缺失值：矩阵、计数、热图和缺失值的树状图
文本分析：了解文本数据的类别（大写、空格）、脚本（拉丁文、西里尔文）和块（ASCII）
文件和图像分析：提取文件大小、创建日期和尺寸,并扫描截断的图像或包含 EXIF 信息的图像

大家可以在 pandas-profiling 的项目 GitHub 页面获取详细使用方法.

机器学习

PyCaret

PyCaret 是 Python 中的一个开源、低代码机器学习库,可自动执行机器学习工作流.It is also an end-to-end machine learning and model management tool,The experimental cycle can be accelerated exponentially,提升工作开发效率.

与其他开源机器学习库相比,PyCaret It has obvious low-code characteristics,What would otherwise require hundreds of lines of code can be done in just a few lines of code,Especially for intensive experimental iterative processes can be greatly accelerated.PyCaret Essentially, it is encapsulated around multiple machine learning libraries and frameworks,包括大家熟悉的 Scikit-Learn、XGBoost、LightGBM、CatBoost、spaCy、Optuna、Hyperopt 和 Ray 等.

大家可以通过 Pycaret 的官方文档,官方GitHub,官方教程 Learn more about usage details.

深度学习

PyTorch Lightning

PyTorch Lightning is an application in deep learning/神经网络的 Python 低代码库,为 PyTorch Provides a high-level interface.

It has a high performance and lightweight architecture,Constructed in a way that separates research from engineering PyTorch 代码,Make deep learning experiments easier to understand and repeat.Use it to easily build scalable deep learning models on distributed hardware.

官网介绍说,PyTorch Lightning is designed so that people can spend more time on research,rather than spending on engineering.大家可以通过 PyTorch Lightning 的官方网站 Learn more about usage details.

Hugging Face Transformers

Hugging Face Transformers 是 Hugging Face A library of open source deep learning tools.借助 Transformers,You can download state-of-the-art pre-trained models very easily and quickly,Apply it in your own scene,Or retrain on your own data.

Because of the large number of pre-trained models officially provided,We can reduce computational costs（Because there is no need to train from scratch）.Rich models cover multiple data types and business sources,包括：

文本：对文本进行分类、信息抽取、问答系统、机器翻译、摘要生成,文本生成（覆盖 100 多种语言）
图像：图像分类、目标检测和图像分割
音频：语音识别和音频分类

PyTorch、TensorFlow 和 JAX are three of the most well-known deep learning libraries,transformers All three frameworks are well supported,It is even possible to train a model with three lines of code in one framework,Load the model in another framework and do inference.

大家可以通过 Hugging Face Transformers 的官方网站和 GitHub Learn more about usage details.