ConvNeXt & EfficientNet¶

Notation Conventions

Parentheses () indicate literal parentheses.
Braces {} are used to bind combinations of options.
The bracket [] indicates an optional clause.
An ellipsis following a comma in brackets [,...] means that the preceding item can be repeated as a comma-separated list
The vertical bar | represents the logic OR.
VALUE represents a regular value.

literal: a fixed or unchangeable value, also known as a Constant.

Each literal has a special data type such as column, in the table.

BUILD MODEL Syntax¶

Use the "BUILD MODEL" statement to develop an AI model. The "BUILD MODEL" statement allows you to train a model using datasets defined with the query_expr that comes after the "AS" clause.

query_statement:
    query_expr

BUILD MODEL (model_name_expression)
USING { ConvNeXt_Tiny | ConvNeXt_Base | EfficientNetV2S | EfficientNetV2M }
OPTIONS (
    expression [ , ...]
    )
AS
(query_expr)

OPTIONS Clause

OPTIONS (
    (image_col=column_name),
    (label_col=column_name),
    [batch_size=VALUE],
    [max_epochs=VALUE],
    [learning_rate=VALUE],
    [input_size=VALUE],
    [color={'RGB'|'GRAY'}],
    [overwrite={True|False}]
    )

The "OPTIONS" clause allows you to change the value of a parameter. The definition of each parameter is as follows.

"image_col": name of column containing the image path to be used for the training (str, default: 'image_path')
"label_col": name of the column containing information about the target value (str, default: 'label')
"batch_size": the size of dataset bundle utilized in a single cycle of training (int, optional, default: 16)
"max_epochs": number of times to train with the training dataset (int, optional, default: 3)
"learning_rate": the learning rate of the model (float, optional default: 1e-3)
"input_size": size of the image to be used for training (int, optional)
"color": color of the image to be used for training (str, optional, 'RGB'|'GRAY', default: 'RGB')
"overwrite": overwrite if a model with the same name exists. If True, the existing model is overwritten with the new model (bool, optional, True|False, default: False)

BUILD MODEL Example

An example "BUILD MODEL" query can be found in Create an Image Classification Model.

%%thanosql
BUILD MODEL my_product_classifier
USING ConvNeXt_Tiny
OPTIONS (
  image_col='image_path',
  label_col='div_l',
  max_epochs=1,
  overwrite=True
  )
AS
SELECT *
FROM product_image_train

FIT MODEL Syntax¶

Use the "FIT MODEL" statement to retrain an AI models. The "FIT MODEL" statement allows you to retrain a model using datasets defined with the query_expr that comes after the "AS" clause. In this case, the label of the data used for retraining should be the same as the label used for the previous training.

query_statement:
    query_expr

FIT MODEL (model_name_expression)
USING (model_name_expression)
OPTIONS (
    expression [ , ...]
    )
AS
(query_expr)

OPTIONS (
    (image_col=column_name),
    (label_col=column_name),
    [batch_size=VALUE],
    [max_epochs=VALUE],
    [learning_rate=VALUE],
    [input_size=VALUE],
    [color={'RGB'|'GRAY'}],
    [overwrite={True|False}]
    )

The "OPTIONS" clause allows you to change the value of a parameter. The definition of each parameter is as follows.

"image_col": name of column containing the image path to be used for the training (str, default: 'image_path')
"label_col": name of the column containing information about the target value (str, default: 'label')
"batch_size": the size of dataset bundle utilized in a single cycle of training (int, optional, default: 16)
"max_epochs": number of times to train with the training dataset (int, optional, default: 3)
"learning_rate": the learning rate of the model (float, optional default: 1e-3)
"input_size": size of the image to be used for training (int, optional)
"color": color of the image to be used for training (str, optional, 'RGB'|'GRAY', default: 'RGB')
"overwrite": overwrite if a model with the same name exists. If True, the existing model is overwritten with the new model (bool, optional, True|False, default: False)

PREDICT Syntax¶

Use the "PREDICT" statement to apply AI models to perform prediction, classification, recommendation, and more. The "PREDICT" statement can preprocess the dataset defined by the query_expr that comes after the "AS" clause.

query_statement:
    query_expr

PREDICT USING (model_name_expression)
OPTIONS (
    expression [ , ...]
    )
AS
(query_expr)

OPTIONS Clause

OPTIONS (
    (image_col=column_name),
    [result_col=column_name],
    [batch_size=VALUE],
    [input_size=VALUE]
    )

The "OPTIONS" clause allows you to change the value of a parameter. The definition of each parameter is as follows.

"image_col": the column containing the image path to be used for prediction (str, default: 'image_path')
"result_col": the column that contains the predicted results (str, optional, default: 'predict_result')
"batch_size": the size of the dataset bundle utilized in a single cycle of prediction (int, optional, default: 16)
"input_size": size of the image to be used for prediction (int, optional)

PREDICT Example

An example "PREDICT" query can be found in Create an Image Classification Model.

%%thanosql
PREDICT USING my_product_classifier
OPTIONS (
    image_col='image_path',
    result_col='predict_result'
    )
AS
SELECT *
FROM product_image_test

EVALUATE Syntax¶

Use the "EVALUATE" statement to evaluate the AI model. The "EVALUATE" statement evaluates a model using the dataset defined by the query_expr that comes after the "AS" clause.

query_statement:
    query_expr

EVALUATE USING (model_name_expression)
OPTIONS (
    expression [ , ...]
    )
AS
(query_expr)

OPTIONS Clause

OPTIONS (
    (image_col=column_name),
    (label_col=column_name),
    [batch_size=VALUE],
    [input_size=VALUE]
    )

The "OPTIONS" clause allows you to change the value of a parameter. The definition of each parameter is as follows.

"image_col": the column containing the image path to be used for evaluation (str, default: 'image_path')
"label_col": the name of the column containing information about the target (str, default: 'label')
"batch_size": the size of dataset bundle utilized in a single cycle of evaluation (int, optional, default: 16)
"input_size": size of the image to be used for evaluation (int, optional)

Last update: 2023-08-09