Train

Train custom models

To train your own custom model, you can choose a pretrained ML model as a based model, and finetuned it with your own data

Prerequisite

GPU is required for training custom models. This feature is currently only available in the distributed version.

Task	Description	Model Name
Text Classification	Use to assign label or class to text, labels such as sentiment or product category	model.text_classification
Question Answering	Use to find answers in a large amount of text	model.question_answering
Recommender	Use to recommend items to user base on collaborative filtering	model.recommender
Summarization	Use to summarize a large amount of text to a short description	model.summarization
Text Generation	Use to generate text from a preceding example	model.text_generation
Translation	Use for translating one language to another, for example English to French	model.translation

Text Classification

Text classification can be use to generate labels from text. For example, let's say you want to train your own custom model to determine the generes of movies, you run the following query to train a model using specific training data from table movie and using the base model text_classification for fine tuning.

To prepare the training dataset, we can run the following query to get the the text and labels

SELECT movie.overview as text, movie.genre as labels FROM mldb.movie as movie

text	labels
Two imprisoned men bond over a number of years, finding solace and eventual redemption through acts of common decency.	Drama
An organized crime dynasty's aging patriarch transfers control of his clandestine empire to his reluctant son.	Crime, Drama
When the menace known as the Joker wreaks havoc and chaos on the people of Gotham, Batman must accept one of the greatest psychological and physical tests of his ability to fight injustice.	Action, Crime, Drama

To train a custom model, we can run the following query

CREATE MODEL model.my_movie_classification_model
FROM (SELECT movie.overview as text, movie.genre as labels FROM mldb.movie) as training_data
WHERE MODEL_NAME = 'text_classification'

model_id	model_name	model_type	training_data	status
my_model_id	movie_genre_classification_model	text_classification	SELECT movie.overview as text, movie.genre as labels FROM mldb.movie	initializing

Recommender

Recommender can be use to predict items that a user will be interested in base on other users with similar interests. For example, let's say we want recommend movies to our users. We can use a database table of movie rating to trained our custom model. To get this sample data, import it from movie rating dataset.

user_id	movie_id	rating
1	31	2.5
1	1029	3

To prepare a training dataset with only the column name user_id and item_id for the model. Movies are rated from 1-5, and since we only need positive rating in our training data, we can run the following query to create the training data for positive reviews higher than 3. To preview the training data we can run the following query

SELECT user_id as user_id, movie_id as item_id FROM mldb.movie_rating WHERE mldb.movie_rating.rating > 3

user_id	item_id
1	1172
1	1339

We can then run the following query to train a model under the name my_movie_recommender_model using our training_data. To indicate that this training is for a recommender model, we will need to pass in the condition MODEL_NAME = 'recommender'

CREATE MODEL model.my_movie_recommender_model
FROM (SELECT user_id as user_id, movie_id as item_id FROM mldb.movie_rating WHERE mldb.movie_rating.rating > 3) as training_data
WHERE MODEL_NAME = 'recommender'

model_id	model_name	model_type	training_data	status
my_model_id	my_movie_recommender_model	recommender	SELECT user_id as user_id, movie_id as item_id FROM mldb.movie_rating WHERE mldb.movie_rating.rating > 3	initializing

The result set will provide the model_id for our custom model. To make predictions for this model, continue at predict recommender model.

Summarization

Summarization can be use to summarize a large amount of text to a short description . For example, let's say you want to train a custom model to summarize the movie description to a few lines, you run the following query to train a model using specific training data from table movie and using the base model summarization for fine tuning.

SELECT movie.overview as summary, movie.description as source FROM mldb.movie

summary	source
Two imprisoned men bond over a number of years, finding solace and eventual redemption through acts of common decency.	......
An organized crime dynasty's aging patriarch transfers control of his clandestine empire to his reluctant son.	......

To train a custom model, we can run the following query

CREATE MODEL model.my_movie_summarization_model
FROM (SELECT movie.overview as summary, movie.description as source FROM mldb.movie) as training_data
WHERE MODEL_NAME = 'summarization'

model_id	model_name	model_type	training_data	status
my_model_id	my_movie_summarization_model	summarization	SELECT movie.overview as summary, movie.details as source FROM mldb.movie	initializing

Translation

Translation model can be use to translate text from one language to another. For example, let's say you want to train a custom model to do french translation for caption in your movies, you run the following query to train a model using specific training data from table movie_captions and using the base model translation for fine tuning.

To get the training data, we can run the following query

SELECT captions.orignal as source, captions.french as target FROM mldb.movie_captions as captions where language = 'french'

source	target
Two imprisoned men bond over a number of years, finding solace and eventual redemption through acts of common decency.	Deux hommes emprisonnés se lient pendant plusieurs années, trouvant un réconfort et une éventuelle rédemption grâce à des actes de décence commune.
An organized crime dynasty's aging patriarch transfers control of his clandestine empire to his reluctant son.	Le patriarche vieillissant d'une dynastie du crime organisé transfère le contrôle de son empire clandestin à son fils réticent.

To train a custom model, we can run the following query

CREATE MODEL model.movie_french_translation_model
FROM (SELECT captions.orignal as source, captions.french as target FROM mldb.movie_captions as captions where language = 'french') as training_data
WHERE MODEL_NAME = 'translation'

model_id	model_name	model_type	training_data	status
my_model_id	my_movie_translation_model	translation	SELECT captions.orignal as source, captions.french as target FROM mldb.movie_captions as captions where language = 'french'	initializing