Volcano engine tool technology sharing: use AI to complete data mining and complete SQL writing with zero threshold

When using BI tools, the question often encountered is: "How can I produce and process data if I don't know SQL? Can I do mining analysis if I don't know the algorithm?"

When professional algorithm teams do data mining, data analysis and visualization will also appear relatively fragmented. Completing algorithm modeling and data analysis work in a streamlined manner is also a good way to improve efficiency.

At the same time, for professional data warehouse teams, data content with the same theme faces the problem of "repeated construction, relatively scattered use and management" - is there any way to produce it at the same time in one task, with the same theme? Datasets with different content? Can the produced data set be used as input to re-participate in data construction?

1. DataWind’s visual modeling capability is here

The BI platform DataWind intelligent data insight launched by the Volcano Engine has launched a new advanced feature-visual modeling.

Users can simplify the complex data processing and modeling process into a clear and easy-to-understand canvas process through visual dragging, pulling, and connecting operations. All types of users can complete data production and processing according to the idea of ​​​​what they think is what they get. Thereby lowering the threshold for data production and acquisition.

Canvas supports the simultaneous construction of multiple groups of canvas processes, and one picture can realize the construction of multiple data modeling tasks, improving the efficiency of data construction and reducing task management costs; in addition, Canvas integrates and encapsulates more than 40 types of data cleaning , feature engineering operators, covering primary to high-level data production capabilities, without the need for coding to complete complex data capabilities.

2. Zero-threshold SQL tools

Data production and processing is the first step to obtain and analyze data.

For non-technical users, there is a certain threshold for using SQL syntax. At the same time, local files cannot be updated regularly, resulting in the dashboard needing to be redone manually every time. The technical manpower required to obtain data often needs to be scheduled, and the timeliness and satisfaction of data acquisition are greatly reduced. Therefore, it is particularly important to use zero-code data construction tools.

Listed below are two typical scenarios of how zero-threshold data processing is applied in work.

2.1 [Scenario 1] What you think is what you get, and the data processing process is completed visually

When product operation iterations are in urgent need of timely input feedback of different data, the data processing process can be abstracted and constructed through visualization The modular drag operator constructs the data processing process.

If you want to obtain the number of orders and order amounts based on date and city granularity, and obtain the city data of the top 10 daily consumption amount data, the operation is as follows:

General data processing process

Visual modeling process

Please technical students pull the detailed data of the order, including order id/order amount/user id/order date city, etc. Set the data through the perspective operation, order date, city, the indicator is the sum of the order amount, the sum of the order id Sort the pivot results by amount, and then write the serial number Use the filter to filter the Top10 data

Select the data source, select the database table or upload a CSV file or connect to LarkSheet Filter the field information you need to use , configure your own defined field names and formats Select the aggregation operator to aggregate and calculate order volume and order amount based on date and city Select the Top value operator and take the Top10 amount Output the data set, which can be applied to Fengshen Draw a chart

2.2 [Scenario 2] Quickly combine multiple tables to easily solve multi-data association calculations

In the data processing process, there are multiple data sources that need to be combined and used. Conventionally, it is difficult and time-consuming to master advanced Vlookup and other algorithms through Excel. At the same time, when the amount of data is large, the computer performance may not be able to complete the combined calculation of the data.

If there are two orders with relatively large amounts of data and a customer attribute information table, the profit amount needs to be calculated based on the bill amount and cost amount, and then the top 100 user order information is taken based on the profit contribution

3. AI data mining is no longer out of reach

When basic data cleaning is no longer able to satisfy data construction and data analysis, AI algorithm support is needed for mining. When data has more hidden value. Algorithm team students may suffer from the inability to work well with visual charts, and cannot produce good data that can be applied quickly; while ordinary users may be directly suppressed by the high threshold of AI code to suppress the emergence of this algorithm - raising demand but fearing demand It is too shallow and the value cannot be well evaluated. At this time, algorithm mining becomes a luxury.

DataWind’s visual modeling encapsulates more than 30 common AI operator capabilities. Users only need to understand the function of the algorithm and configure the input and training goals of the algorithm operator through configuration to complete model training. , quickly obtain prediction results based on other configured data content. ​

#The following will take two typical scenarios as examples, How to complete data mining without knowing how to write Python.

​

3.1 [Elementary] You can do data mining even if you don’t know Python

Users’ daily work basically does not involve writing Python, but there are demand scenarios for data mining. He needs to do customer intention mining based on existing high-intention customer samples. At this point, the data mining process can be built through visual modeling:

Drag in the sample data and all data as data input.
1. Drag into the classification algorithm, such as XGB algorithm for model training.
2. Drag in the prediction operator and build the relationship between the model and all the data for prediction.
3. The actual data and predicted results are combined with the output data set to analyze the intention distribution of all user data.

3.2 [Advanced] You can build complex algorithm models without writing Python

Users need to build a model based on existing data User repurchase model. During model building, it is necessary to use gradient boosting tree to build a prediction model after data cleaning and format conversion. At this time, the repurchase model process can be built based on visual modeling:

Merge rows: Merge the output data tables of n operators (rectangles in the figure) into one overall data table based on consistent headers. If the user's sales data does not add or delete new attributes, no changes are needed here.
1. Missing value replacement: When there is a null value (null) in the attribute column, it will affect subsequent model calculations. Use the replace missing value operator to replace the null value with the specified default value. When the user's sales data does not add or delete new attributes. No changes are needed here.
2. one-hot encoding: Text type attributes cannot be directly used in model training and need to be encoded into a numeric vector by one_hot. For example:

Gradient Boosting Tree: Responsible for fitting training data and outputting a model that can be used for prediction (parameters not marked in the figure do not need to be modified by maintainers):

Aggregation_1: Remove duplicates in the prediction data and take the maximum probability.
5. Extract fields: Extract the necessary label and probability value output.

4. Multi-scenario and multi-task construction, management is no longer decentralized

As a data analyst, you will also have a lot of daily work to build data sets and build data dashboards. But usually the bottom table obtained from the data warehouse will be a wide table. On this basis, different data set tasks are built according to different scenario requirements.

In subsequent use, we often encounter more and more similar data sets, but the specific logic cannot be well compared and confirmed. At this time, it would be great if all the data set logic is configured and generated in one data set, and each data set can be judged and defined through the task process.

For this scenario, DataWind’s visual modeling capabilities can also be completed very well. The visual modeling function supports a single data set to be processed by multiple logic processes at the same time to generate multiple data sets. Take the processing of order data and user data as an example:

1. If a user wants to see the order statistics, then the data processing process of the order statistics data set can be built.
2. There are users who want to see detailed data, but the detail fields need to be processed and cleaned. In this case, the processing flow of the order detail table data set can be constructed.
3. Some users want to combine user attributes to count the user's order distribution, then build a multi-table association combined with indicator aggregation to generate a complete user order statistical data set.
4. The same logic can generate user order details data set under multi-table association.

Thus, four data sets are generated through one task and two data inputs. The four data sets can construct a data subject domain, and subsequent related data can be used from this point on. The data set output by the task is used.

5. About us

Volcano Engine Intelligent Data Insight DataWind is an enhanced platform that supports self-service analysis at the detailed level of big data. ABI platform. From data access, data integration, to query and analysis, it is finally presented to business users in the form of visual data portals, digital large screens, and management cockpits, allowing data to exert value.

General data processing process

Visual modeling process

Requires two After the order data is opened, the Copy data is merged into a file Use VloopUp to find the user data in the order and the user data in the customer, and then combine the two data to generate new data Use a pivot table to calculate the user bill amount and cost amount, and then calculate the profit amount ## Get TopN customer information sorted by profit amount

## You can upload CSV files/LaskSheet to build data input Then you can merge the March/April order data into one data Connect the customer information attribute table and bind the customer Attribute information Select aggregation to calculate the bill amount and cost amount according to the customer specific number ## Select the calculation column to calculate the profit amount based on the bill amount and cost amount Get the TopN customer information sorted according to the profit amount

The above is the detailed content of Volcano engine tool technology sharing: use AI to complete data mining and complete SQL writing with zero threshold. For more information, please follow other related articles on the PHP Chinese website!