AlgorithmSpec

Updated at:

Parameter

Type

Description

Example

object

The algorithm definition.

OutputChannels

array

The list of output channels for the algorithm.

Channel

An output channel for the algorithm.

HyperParameters

array

The list of hyperparameter definitions.

HyperParameterDefinition

A hyperparameter definition.

JobType

string

The job type.

TFJob

Command

array

The list of commands to run the training task.

string

A command to run the training task.

python train.py

MetricDefinitions

array

The list of metric definitions for the training task.

MetricDefinition

A metric definition for the training task.

InputChannels

array

The list of input channels for the algorithm.

Channel

An input channel for the algorithm.

SupportsDistributedTraining

boolean

Indicates whether distributed training is supported.

true

Image

string

The training runtime image.

registry.cn-shanghai.aliyuncs.com/pai-training/kmeans:v1.0.0

SupportedInstanceTypes

array

The list of supported elastic computing instance types for training.

string

A supported elastic computing instance type for training.

ecs.g6.2xlarge

ComputeResource

object

The compute resource definition.

Policy

object

The compute resource configuration policy for the algorithm.

Version

string

The version of the configuration policy.

V1

Value

string

The policy content, serialized from a JSON array.

[{\"MinTotalCount\":1,\"MaxTotalCount\":4,\"RolePolicies\":{\"chief\":{\"Count\":1},\"ps\":{\"Count\":1},\"worker\":{\"Percentage\":100}}},{\"MinTotalCount\":5,\"MaxTotalCount\":10,\"RolePolicies\":{\"chief\":{\"Count\":1},\"ps\":{\"Percentage\":20},\"worker\":{\"Percentage\":80}}},{\"MinTotalCount\":11,\"RolePolicies\":{\"chief\":{\"Count\":1},\"ps\":{\"Percentage\":40},\"worker\":{\"Percentage\":60}}}]

CodeDir Location

The custom code configuration.

Customization

object

The custom configuration.

CodeDir

boolean

Indicates whether to use custom mounted code.

true

ResourceRequirements

array

The resource requirements.

ConditionExpression

Expression of resource requirements.

ProgressDefinitions

object

You can use this feature to monitor the training progress.

OverallProgress

object

Monitors training progress.

Description

string

The description of the monitoring definition.

training progress

Regex

string

The regular expression for monitoring.

^[0-9]+([.][0-9]+){0,1}$

RemainingTime

object

The definition for monitoring the remaining training time.

Description

string

The description of the monitoring definition.

training remaining time

Regex

string

The regular expression for monitoring.

^[0-9]+([.][0-9]+){0,1}$