​
MindsDB NLP Supported Tasks

There are four main NLP tasks currently supported by MindsDB:

  • Text Classification
  • Zero-Shot Classification
  • Translation
  • Summarization

Currently, MindsDB’s NLP engine is powered by Hugging Face and OpenAI. But we plan to expand to other NLP options in the future, so stay tuned!

The MindsDB’s Hugging Face engine is extensible. We are actively working on adding more tasks and models. If you have a specific task or model in mind, please let us know in the MindsDB Community.

​
MindsDB NLP Tested Models

Completes the task of assigning a label to a text. For example, you can use it to classify a movie review as positive or negative.

Supported Models

Completes the task of assigning a label to a text without training on the labels. For example, you can use it to classify a movie review as positive or negative without training on positive or negative labels.

Supported Models

-BART

Completes the task of translating a text from one language to another. For example, you can use it to translate a text from English to French.

Supported Models

Completes the task of summarizing a text. For example, you can use it to summarize an abstract into a title.

Supported Models

The Hugging Face models are used to perform these tasks. Keep in mind that usually there is more than one model for each task, so you can choose the one that suits you best.

​
How to Bring the Hugging Face Model to MindsDB

We use the CREATE MODEL statement to bring the Hugging Face models to MindsDB.

Generally, it looks like this:

CREATE MODEL project_name.predictor_name        -- AI TABLE TO STORE THE MODEL
PREDICT target_column                           -- NAME OF THE COLUMN TO STORE PREDICTED VALUES
USING
  engine = 'huggingface',                       -- USING THE HUGGING FACE ENGINE
  task = 'task',                                -- TASK OF CLASSIFYING TEXT (OPTIONAL)
  model_name = 'model_name_from_hugging_face',  -- MODEL NAME UNDER THE HUGGING FACE MODEL HUB
  input_column = 'input_column',                -- COLUMN NAME OF THE INPUT DATA
  labels = ['label', 'label_1'];                -- ARRAY OF LABELS

Where:

ExpressionsDescription
project_nameName of the project where the model is created. By default, the mindsdb project is used.
predictor_nameName of the model to be created.
target_columnColumn to store the predicted values.
engineOptional. You can provide an ML engine, based on which the model is created.
taskOptional. It is relative to the Hugging Face task tag.
model_nameModel name from the Hugging Face model hub.
input_columnName of the column that has the input data, especially important for batch predictions using JOIN.
labelsDepending on the model. Usually used for Zero-Shot Classification models.

For more examples and explanations, visit our doc page on Hugging Face.

​
Example using SQL

Let’s go through a Spam Classification example to understand better how to link Hugging Face models and bring them to MindsDB as AI tables.

Using Local Installation of MindsDB

Please note that if you use a local installation of MindsDB, instead of MindsDB Cloud, you should install transformers==4.21.0 to be able to use the Hugging Face models.

CREATE MODEL mindsdb.spam_classifier                           
PREDICT PRED                           
USING
  engine = 'huggingface',              
  task = 'text-classification',        
  model_name = 'mrm8488/bert-tiny-finetuned-sms-spam-detection', 
  input_column = 'text_spammy',        
  labels = ['ham', 'spam'];

Where:

ExpressionsValues
project_namemindsdb
predictor_namespam_classifier
target_columnPRED
enginehuggingface
tasktext-classification
model_namemrm8488/bert-tiny-finetuned-sms-spam-detection
input_columntext_spammy
labels[‘ham’, ‘spam’]

On execution, we get:

Query successfully completed

Before querying for predictions, we should verify the status of the spam_classifier model.

DESCRIBE spam_classifier;

On execution, we get:

+---------------+-------+--------+--------+-------+-------------+---------------+------+-----------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
|NAME           |PROJECT|STATUS  |ACCURACY|PREDICT|UPDATE_STATUS|MINDSDB_VERSION|ERROR |SELECT_DATA_QUERY|TRAINING_OPTIONS                                                                                                                                                                                               |
+---------------+-------+--------+--------+-------+-------------+---------------+------+-----------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
|spam_classifier|mindsdb|complete|[NULL]  |PRED   |up_to_date   |22.10.2.1      |[NULL]|[NULL]           |{'target': 'PRED', 'using': {'engine': 'huggingface', 'task': 'text-classification', 'model_name': 'mrm8488/bert-tiny-finetuned-sms-spam-detection', 'input_column': 'text_spammy', 'labels': ['ham', 'spam']}}|
+---------------+-------+--------+--------+-------+-------------+---------------+------+-----------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+

Once the status is complete, we can query for predictions.

SELECT h.PRED, h.PRED_explain, t.text_spammy AS input_text
FROM example_db.demo_data.hf_test AS t
JOIN mindsdb.spam_classifier AS h;

On execution, we get:

+----+---------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
|PRED|PRED_explain                                             |input_text                                                                                                                                                       |
+----+---------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+
|spam|{'spam': 0.9051626920700073, 'ham': 0.09483727067708969} |Free entry in 2 a wkly comp to win FA Cup final tkts 21st May 2005. Text FA to 87121 to receive entry question(std txt rate)T&C's apply 08452810075over18's      |
|ham |{'ham': 0.9380123615264893, 'spam': 0.061987683176994324}|Nah I don't think he goes to usf, he lives around here though                                                                                                    |
|spam|{'spam': 0.9064534902572632, 'ham': 0.09354648739099503} |WINNER!! As a valued network customer you have been selected to receivea £900 prize reward! To claim call 09061701461. Claim code KL341. Valid 12 hours only.    |
+----+---------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------+

For the full library of supported examples please go here.

​
Example using MQL

Let’s go through a Sentiment Classification example, but this time we’ll use a Mongo database.

Using Local Installation of MindsDB

Please note that if you use a local installation of MindsDB, instead of MindsDB Cloud, you should install transformers==4.21.0 to be able to use the Hugging Face models.

We have a sample Mongo database that you can connect to your MindsDB Cloud account by running this command in Mongo Shell:

use mindsdb

Followed by:

db.databases.insertOne({
    'name': 'mongo_test_db', 
    'engine': 'mongodb',
    'connection_args': {
        "host": "mongodb+srv://admin:201287aA@cluster0.myfdu.mongodb.net/admin?authSource=admin&replicaSet=atlas-5koz1i-shard-0&readPreference=primary&appname=MongoDB%20Compass&ssl=true",
        "database": "test_data"
        }   
})

We use this sample database throughout the example.

The next step is to create a connection between Mongo and MindsDB. Follow the instructions to connect MindsDB via Mongo Compass or Mongo Shell.

Now, we are ready to create a Hugging Face model.

db.models.insertOne({
    name: 'sentiment_classifier',
    predict: 'sentiment',
    training_options: {
            engine: 'huggingface',
            task: 'text-classification',
            model_name: 'cardiffnlp/twitter-roberta-base-sentiment',
            input_column: 'comment',
            labels: ['negative','neutral','positive']
           }
})

On execution, we get:

{ acknowledged: true,
  insertedId: ObjectId("63c00c704d444a0b83808420") }

We can check its status using this command:

db.getCollection('models').find({'name': 'sentiment_classifier'})

On execution, we get:

{ NAME: 'sentiment_classifier_hf',
  PROJECT: 'mindsdb',
  VERSION: 1,
  STATUS: 'complete',
  ACCURACY: null,
  PREDICT: 'sentiment',
  UPDATE_STATUS: 'up_to_date',
  MINDSDB_VERSION: '22.11.4.3',
  ERROR: null,
  SELECT_DATA_QUERY: null,
  TRAINING_OPTIONS: '{\'target\': \'sentiment\', \'using\': {\'task\': \'text-classification\', \'model_name\': \'cardiffnlp/twitter-roberta-base-sentiment\', \'input_column\': \'comment\', \'labels\': [\'negative\', \'neutral\', \'positive\']}}',
  TAG: null }

Once the status is complete, we can query for predictions.

Here is how to query for a single prediction:

db.sentiment_classifier.find({comment: 'It is really easy to do NLP with MindsDB'})

On execution, we get:

{ sentiment: 'positive',
  sentiment_explain: 
   { positive: 0.9350261688232422,
     neutral: 0.06265384703874588,
     negative: 0.0023200225550681353 },
  comment: 'It is really easy to do NLP with MindsDB' }

You can also query for batch predictions. Here we use the mongo_test_db database connected earlier in this example.

db.sentiment_classifier.find(
    {'collection': 'mongo_test_db.user_comments'},
    {'sentiment_classifier.sentiment': 'sentiment',
     'user_comments.comment': 'comment'
    }
)

On execution, we get:

{ sentiment: 'positive', comment: 'I love pizza' }
{ sentiment: 'negative', comment: 'I hate dancing' }
{ sentiment: 'neutral', comment: 'Baking is not a big deal' }

For the full library of supported examples please go here.

​
What’s Next?

Have fun while trying it out yourself!

If this tutorial was helpful, please give us a GitHub star here.