Introduction to Alibaba Cloud Machine Learning PAI
Summer
Joe Hisaishi-"Kikujiro no Natsu" Original Soundtrack
Machine learning can be roughly divided into three categories:
Supervised learning: refers to that each sample has a corresponding expected value, and the model is built to complete the mapping from the input feature vector to the target value. Typical examples are regression and classification problems, such as logic Regression, random forest, decision tree.
Unsupervised learning: refers to that there is no target value in all samples, and it is expected to discover some potential laws from the data itself, such as some simple clustering K-means, DBSCAN, etc.
Reinforcement learning: Relatively complex, it refers to a system that continuously interacts with the external environment, obtains external feedback, and then determines its own behavior to achieve the optimization of long-term goals. The typical case is Alpha Go, or unmanned driving.
Advertising
Machine Learning
Author: Zhou Zhihua
Dangdang
Alibaba Cloud machine learning platform is built on Alibaba Cloud, which is a machine learning platform that integrates data processing, modeling, offline prediction, and online prediction. Alibaba Cloud machine learning encapsulates the mature algorithms in the Alibaba Group and provides a simpler operation experience for machine learning users.
The intelligent platform is mainly divided into three layers: the
first layer: the web UI layer; the
second layer: the machine learning algorithm layer; the
third layer: the Maxcompute platform layer.
The Web UI interface is mainly composed of the following areas:
Main functional area: Displays the names of various components.
Canvas area: The user can drag and drop the corresponding components onto the canvas with the mouse to form a directed workflow to complete the processing from data to data. Then to a series of data mining tasks such as modeling.
Attribute area: This area can set the parameter information in the component.
Alibaba Cloud machine learning infrastructure:
Infrastructure layer: CPU computing cluster
computing framework layer: Including MaxReduce, SQL, MPI and other computing methods, the distributed computing framework mainly performs parallel computing and distribution tasks.
Machine Learning PAI advantages: 1. 2. rich deep learning algorithm (GPU) 3. visual interface
4. The one-stop service
at the same time, the depth of PAI learning algorithm support frame: 1.Tensorflow 2.Caffe 3.MXNet
Complete the following basic tasks on the machine learning platform:
open machine learning services,
data preparation,
data preprocessing,
data visualization
algorithm modeling
model evaluation. In the
data preparation phase, the bottom layer of the machine learning platform supports two data sources, one is MaxCompute to store data and the other One is OSS storing data.
Note: Use MaxCompute as storage. It is recommended to use the machine learning IDE environment to upload when the data is less than 20MB, and use the command line tool to upload when the data is larger than 20MB.
Open machine learning PAI, and create a project, pay attention to the region you choose when you open it.
2. Data preparation, enter the machine learning platform, click the data source, create a table.
Upload data to Maxcompute from IDE
OSS upload data to create new blank data
- After the data preparation is complete, click the component, drag the SQL script, type conversion, and normalization component under the tools and data preprocessing folder to the canvas, and splice it into the following experiment.
- data visualization
- In the Machine Learning -> Binary Classification folder, drag the Logistic Regression Binary Classification component into the canvas.
On the field setting tab on the right, set the target column to ifhealth, and select all the columns except the target column for the training feature column, and splice and run, as shown in the figure below.
6. Model evaluation
In the machine learning -> evaluation folder, drag the two-class evaluation component into the canvas. Set the tab in the field on the right side of the canvas, set the original label column name to ifhealth, and connect the corresponding component stream and data stream.
Click Run. After completion, right-click the two-category evaluation component, select View Evaluation Report, and click the chart tab to get the ROC curve of the LR model trained under different parameters, as shown in the figure below.
I like to remember to come to a 
"Nezhatou"-play with small trends