Updated on 2023-10-23 GMT+08:00

GS_OPT_MODEL

GS_OPT_MODEL is a data table used when the AI engine is enabled to predict the planned time. It records the configurations, training results, features, corresponding system functions, and training history of machine learning models.

Table 1 GS_OPT_MODEL columns

Name

Type

Description

oid

oid

Database object ID

template_name

name

Template name of the machine learning model, which determines the interfaces invoked for training and prediction. Currently, only rlstm is implemented.

model_name

name

Model name. Each model corresponds to a set of parameters, training logs, and model coefficients in the AI engine online learning process. The name must be unique.

datname

name

Name of the database served by the model. Each model is specific to a single database. This parameter determines data used for training.

ip

name

IP address of the host where the AI engine is deployed

port

integer

Listening port number of the AI engine

max_epoch

integer

Maximum number of iterations in an epoch

learning_rate

real

Learning rate of model training. The default value 1 is recommended.

dim_red

real

Number of model feature dimensions whose retention is reduced

hidden_units

integer

Number of neurons in the model's hidden layer. If the model cannot be converged for a long time, increase the value of this parameter.

batch_size

integer

Size of a batch in each iteration. It is recommended that the size be greater than or equal to the total training data volume to accelerate model convergence.

feature_size

integer

Length of the model feature, which is used to trigger retraining. This parameter is automatically updated after model training and does not need to be specified.

available

boolean

Whether the model is converged. This parameter does not need to be specified.

Is_training

boolean

Whether the model is being trained. This parameter does not need to be specified.

label

"char"[]

Target task of the model.

  • S: startup time
  • T: total time
  • R: rows
  • M: peak memory

Currently, {S, T} or {R} is recommended due to model performance restrictions.

max

bigint[]

Maximum value of each task label of the model, which is used to trigger retraining. This parameter does not need to be specified.

acc

real[]

Accuracy of each model task. This parameter does not need to be specified.

description

text

Model comment