Platform For AI:Billing of Machine Learning Designer

Billing description

When you use Machine Learning Designer algorithm components, you may incur costs for PAI algorithm components, PAI-DLC, or MaxCompute. This topic uses PAI algorithm component costs as an example. For more information about other costs, see DLC billing and MaxCompute billing. All prices in this topic are for reference only. The actual price is displayed on your bill.

Billable items

Billing methods

When you use Machine Learning Designer algorithm components, the unit price varies by algorithm type. The billing is calculated as follows:

• Billing formula: Bill amount = Number of billable hours × Unit price

  Number of billable hours = max(Number of CPU cores × Duration (h), Memory (GB) × Duration (h) / 4)

• Billing duration: The billing duration starts when the component starts and ends when the component stops.

• Billing example: Assume that you use a data analytics algorithm that consumes 2 CPU cores and 5 GB of memory for 1 hour and 30 minutes. The number of billable hours is calculated as follows: max(2 × (1 + 30/60), 5 × (1 + 30/60) / 4) = 3. The bill amount is calculated as follows: Bill amount = Number of billable hours × Unit price = 3 × 0.21 = 0.63 USD. The unit prices for different algorithm types are listed in the following table:

  Algorithm type	Unit price (USD/billable hour)	Algorithm type description
  Data pre-processing (data_manipulation)	0.16	Includes data pre-processing and feature engineering algorithm components.
  Data analytics (data_analysis)	0.21	Includes statistical analysis, machine learning, time series, network analysis, and finance algorithm components. Common algorithms include binning and swing_rec_ext.
  Text analytics (text_analysis)	0.27	Includes text analytics algorithm components.
  Deep learning algorithm (deep_learning)	0.16	Deep learning components such as the CPU version of the TensorFlow framework and EasyRec series algorithms.

  Note

  • If you enable MaxCompute when you activate PAI, you are charged for MaxCompute resources when you use the SQL Script, Join, Union, and Filtering and Mapping components. For more information, see MaxCompute billing.

  • If you enable Flink when you activate PAI and select Flink as the compute resource, you are charged for Flink resources when you use Alink components. For more information, see Flink billing.

Billing example

Typically, an experiment in Machine Learning Designer consists of multiple algorithm components, and each algorithm component consists of multiple subtasks. Therefore, to calculate the cost of an experiment, you must first calculate the cost of the subtasks for each algorithm component and then sum the costs to obtain the total cost.

1. Identify the category of the algorithm component.

   1. Log on to the PAI console.

   2. In the upper-left corner, select a region.

   3. In the left navigation pane, click Workspaces. On the resulting page, click the name of the target workspace.

   4. In the navigation pane on the left, choose Model Development and Training > Machine Learning Designer.

   5. On the Workflows page, select the target workflow and click Go to Workflow.

   6. In the component list, find the PLDA component. It belongs to the text analytics category and is priced at 0.27 USD per billable hour.

      

2. View the resources consumed by the job.

   1. On the experiment canvas, right-click the PLDA component.

   2. From the shortcut menu, select View Log.

   3. On the Log-PLDA page, each blue link corresponds to a subtask. You can click a link.

      

   4. On the LogView page, click the SourceXML tab.

   5. Under TaskPlan, view the CPU and Memory.

      • Divide the value of CPU by 100 to calculate the number of CPU cores used. This job uses 1 CPU core.

      • The unit of Memory is MB. This job uses 1 GB of memory.

   6. On the LogView page, click the Job Details tab.

   7. Click the task object on the AlgoTask_0_0 tab. In the dialog box that appears, click the Terminated tab. The latency shows the runtime of each job.

      This subtask has 49 jobs, and each job runs for about 26 seconds.

3. Calculate the cost of the subtask.

   1. Number of billable hours for the subtask = max(Number of CPU cores × Duration (h), Memory (GB) × Duration (h) / 4) = max(49 × 1 × (26/3600), 49 × 1 × (26/3600) / 4) ≈ 0.35 billable hours

   2. Cost of the subtask = Number of billable hours for the subtask × Unit price = 0.35 × 0.27 = 0.095 USD

4. Calculate the costs of all subtasks in the PLDA component and sum them to obtain the total cost of the PLDA component.

5. Repeat the preceding steps to calculate the costs of all components in the experiment and sum them to obtain the total cost of the experiment.