Xiaoyu's in-depth product evaluation: Alibaba Cloud machine learning platform PAI, it turns out that machine learning can be so easy.

0 Preface

With the blessing of artificial intelligence, machine learning, deep learning, algorithms, model building, etc.,
have been laid out by many large companies.
Early entry, early layout, and early output are necessary to occupy a favorable position in the current AI era. condition.
Of course, the launch of Alibaba Cloud's machine learning platform PAI product is expected.
Moreover, I am also fortunate to participate in the evaluation of the machine learning platform PAI. For me, it is growth and learning.
So, next, with a learning attitude, I will make some output in the use and feeling of the product.

1. Unboxing experience

To use it, it must be out of the box.
Here I directly click the "Activate Now" → "Activate for free and create a default workspace" button to enter today's topic.

Click the "Confirm activation..." button to live up to expectations, and a prompt pop-up window will pop up to authorize.

This time I clicked the "Direct Open" button to see what happened.

It's perfect, because I have resource arrears, so I must not be activated to avoid being "free prostitution".

To change our thinking, let's go to "Authorization",

Seeing this prompt means that the authorization is successful and can be used.

Here is a question :
After the authorization is successful, I closed all the previous pages, how to return at this time?

After my unremitting efforts, finally opened successfully.

Looking at the interface, is it familiar and unfamiliar?
Don't worry, let's experience the satisfaction brought by the PAI platform a little bit...

2. Workspace

2.1 Workspace Details

The workspace details page can display AI assets, which is very convenient.
Of course, what is essential for this is the operation steps, that is, how to get started quickly.

Cloud native development scenario, as shown in the following figure:

The best practice of AI + big data, as shown in the figure below:

2.2 Cloud Native Development Scenarios

The start button for the direct dataset ,

Directly enter the dataset management menu page, and the overall layout of the page is very simple and clear.

Moreover, you can display the field information of the list according to your own needs, which is very good.

2.3 Best Practices of AI + Big Data

I directly click the start button of the "Visual Modeling" module of AI+Big Data Best Practices ,

Jump directly to the " Visual Modeling " menu, which is very convenient.
This saves the time of searching menu by menu, so this flow chart is very practical.

3. Model development and training

3.1 Visual modeling

3.1.1 New

Workflow list tab, we first create a new workflow,

The default page is displayed, as shown below :

I just fill in some information and save it

After saving, the list is as follows

The new "_" workflow has been generated in the root directory. And I have created several new workflows one after another, and there are no problems.

3.1.2 Import

I'll import the workflow again and try it out.

The json file has been uploaded, as shown below

The file is not uploaded, as shown below

6 Questions & Questions:

• ①. The workflow name here will automatically add "_copy" according to the name of the json file you uploaded.
  I don't know if this is done on purpose or for other reasons?

• ②. The uploaded json file does not verify the content, but only verifies that the json format is correct.

• ③. If the file is not uploaded, click the OK button, it will prompt that the "workflow name" has not been filled in. In fact, if you look at it again, the workflow name is grayed out. How to fill it in is confusing? ?

• ④. Modify the workflow name of the imported file. I just remove _, and an error will be reported and cannot be saved (as shown below)...
  

• ⑤. When the name of the workflow is 100 characters, the list is incomplete and cannot be displayed when the mouse hovers over it. Only by entering the work basic information page can all be displayed, as shown in the figure below
  

• ⑥. There is no repetition restriction on the name of the workflow, so, for the case of ⑤, it is not clear whether it is in the list.
  

3.1.3 Publish as Workspace Visible

Select the newly created work workflow folder, click OK, 1 This workflow changes the folder correctly, no problem.

3.1.4 Query

The query box here is very simple, it is a prompt: please enter

I entered the workflow folder name and workflow ID, but I can't find it

Only the workflow name can be queried

Suggestion:
If the little developer doesn't find it troublesome, let's complete the copy of the input box, and these two words are not bad.

3.1.5 Cloning

• 1. Create a new workflow and clone the page, like this.
  

• 2. Import workflow, clone
  

question:

• If you click the OK button without modifying the workflow name, it cannot be saved, and it will prompt that the workflow name is not standardized...

• In fact, when uploading a file, you need to verify the file naming convention, otherwise, the clone will prompt that the workflow file name is not standardized, that is, as shown in the following figure
  

3.2 Enter the workflow

3.2.1 Overall display of the workflow page

Enter the workflow page, the overall style is as follows:

• The workflow shown on the left is the use function;
• The basic information of the workflow is displayed on the right;
• The content of the workflow is displayed in the middle;

3.2.2 Running

Click the run button, it is very powerful, and an error is reported

because the content is empty.

After I adjusted again, the page became like this

Then run it again.
One thing to note here:
if your working node uses other resources, you need to add it before running it, otherwise an error will be reported, as shown in the figure below:

3.2.3 Tool Model

Powerful features on the left, we can see.

Here is a very convenient point. If you don't know how to use the current model/tool, a link to the document will be provided directly. This is very user-friendly.

3.3 Prefabricated formwork

In the prefabricated templates, you can see that there are very comprehensive templates for you to use.

3.3.1 Prefabricated Template - Create

Here, I also create a haze weather forecast

Created and that's it.

4. Summary

As for other models, I also used them one by one.
In order to save resources, I will not show screenshots one by one.

In the overall use process, I still feel powerful, convenient and detailed.

Although the entire platform still has a little bit of flaws, but the flaws do not hide the goodness.
As the saying goes, the imperfect product is the most perfect.
Therefore, overall, there is no problem with the use of the machine learning PAI platform.
And it is also very consistent with the theme of the product: to consolidate the engineering platform and jointly build an open source ecological AI.

At the same time, I also hope that Alibaba Cloud will uphold more missions to improve its technical capabilities and launch more convenient products that better meet the needs of enterprises and individuals.