Skip to content

Trainable Part of Speech Tagger (POS), Sentiment Classifier with BERT/USE/ELECTRA sentence embeddings in 1 Line of code! Latest NLU Release 1.0.5
Compare
Choose a tag to compare
@C-K-Loan C-K-Loan released this 15 Dec 02:57
· 1304 commits to master since this release
v1.0.5
73cc744
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

NLU 1.0.5 Release Notes

We are glad to announce NLU 1.0.5 has been released!
This release comes with a trainable Sentiment classifier and a Trainable Part of Speech (POS) models!
These Neural Network Architectures achieve the state of the art (SOTA) on most binary Sentiment analysis and Part of Speech Tagging tasks!
You can train the Sentiment Model on any of the 100+ Sentence Embeddings which include BERT, ELECTRA, USE, Multi Lingual BERT Sentence Embeddings and many more!
Leverage this and achieve the state of the art in any of your datasets, all of this in just 1 line of Python code

NLU 1.0.5 New Features

  • Trainable Sentiment DL classifier
  • Trainable POS

NLU 1.0.5 New Notebooks and Tutorials

Sentiment Classifier Training

Sentiment Classification Training Demo

To train the Binary Sentiment classifier model, you must pass a dataframe with a 'text' column and a 'y' column for the label.

By default Universal Sentence Encoder Embeddings (USE) are used as sentence embeddings.

fitted_pipe = nlu.load('train.sentiment').fit(train_df)
preds = fitted_pipe.predict(train_df)

If you add a nlu sentence embeddings reference, before the train reference, NLU will use that Sentence embeddings instead of the default USE.

#Train NER on BERT sentence embeddings
fitted_pipe = nlu.load('embed_sentence.bert train.classifier').fit(train_df)
preds = fitted_pipe.predict(train_df)
#Train NER on ELECTRA sentence embeddings
fitted_pipe = nlu.load('embed_sentence.electra train.classifier').fit(train_df)
preds = fitted_pipe.predict(train_df)

Part Of Speech Tagger Training

Your dataset must be in the form of universal dependencies Universal Dependencies.
You must configure the dataset_path in the fit() method to point to the universal dependencies you wish to train on.
You can configure the delimiter via the label_seperator parameter
[POS training demo]](https://colab.research.google.com/drive/1CZqHQmrxkDf7y3rQHVjO-97tCnpUXu_3?usp=sharing)

fitted_pipe = nlu.load('train.pos').fit(dataset_path=train_path, label_seperator=',')
preds = fitted_pipe.predict(train_df)

NLU 1.0.5 Installation changes

Starting from version 1.0.5 NLU will not automatically install pyspark for users anymore.
This enables easier customizing the Pyspark version which makes it easier to use in various cluster environments.

To install NLU from now on, please run

pip install nlu pyspark==2.4.7

or install any pyspark>=2.4.0 with pyspark<3

NLU 1.0.5 Improvements

  • Improved Databricks path handling for loading and storing models.
Assets 3
Loading