Overview | Tutorials | Examples | Installation | FAQ | How to Cite
- 2021-03-10:
-
ktrain v0.26.x is released and now supports
transformers>=4.0.0
.
Note that,transformers>=4.0.0
included a complete reogranization of the module's structure. This means that, if you saved a transformers-basedPredictor
(e.g., DistilBERT) in an older version of ktrain and transformers, you will need to either generate a newtf_model.preproc
file or manually edit the existingtf_model.preproc
file before loading thepredictor
in the latest versions of ktrain and transformers.
For instance, suppose you trained a DistilBERT model and saved the resultant predictor using an older version of ktrain with:predictor.save('/tmp/my_predictor/')
. After upgrading to the newest version of ktrain, you will find thatktrain.load_predictor('/tmp/my_predictor
) will throw an error unless you follow one of the two approaches below:Approach 1: Manually edit
tf_model.preproc
file:
Opentf_model.preproc
with an editor like vim and edit it to replace old module locations with new module locations (example changes for a DistilBERT model shown below):# change transformers.configuration_distilbert to transformers.models.distilbert.configuration_distilbert # change transformers.modeling_tf_auto to transformers.models.auto.modeling_tf_auto # change transformers.tokenization_auto to transformers.models.auto.tokenization_auto
The above was confirmed to work using the vim editor on Linux.
Approach 2: Re-generate
tf_model.preproc
file:# Step 1: Re-create a Preprocessor instance # NOTES: # 1. If training set is large, you can use a sample containing at least one example for each class # 2. Labels must be in same format as you originally used # 3. If original training set is not easily accessible, set preproc.preprocess_train_called=True # below instead of invoking preproc.preprocess_train(x_train, y_train) preproc = text.Transformer(MODEL_NAME, maxlen=500, class_names=class_names) trn = preproc.preprocess_train(x_train, y_train) # Step 2: load the transformers model from predictor folder from transformers import * model = TFAutoModelForSequenceClassification.from_pretrained('/tmp/my_predictor/') # Step 3: re-create/re-save Predictor predictor = ktrain.get_predictor(model, preproc) predictor.save('/tmp/my_new_predictor')
-
If you're using PyTorch 1.8 or above with ktrain, you will need to upgrade to
ktrain>=0.26.0
. If you're usingktrain<0.26.0
, then you will have to downgrade PyTorch with:pip install torch==1.7.1
.
-
- 2020-11-08:
- ktrain v0.25.x is released and includes out-of-the-box support for text extraction via the textract package . This, for example,
can be used in the
SimpleQA.index_from_folder
method to perform Question-Answering on large collections of PDFs, MS Word documents, or PowerPoint files. See the Question-Answering example notebook for more information.
- ktrain v0.25.x is released and includes out-of-the-box support for text extraction via the textract package . This, for example,
can be used in the
# End-to-End Question-Answering in ktrain
# index documents of different types into a built-in search engine
from ktrain import text
INDEXDIR = '/tmp/myindex'
text.SimpleQA.initialize_index(INDEXDIR)
corpus_path = '/my/folder/of/documents' # contains .pdf, .docx, .pptx files in addition to .txt files
text.SimpleQA.index_from_folder(corpus_path, INDEXDIR, use_text_extraction=True, # enable text extraction
multisegment=True, procs=4, # these args speed up indexing
breakup_docs=True) # this slows indexing but speeds up answer retrieval
# ask questions (setting higher batch size can further speed up answer retrieval)
qa = text.SimpleQA(INDEXDIR)
answers = qa.ask('What is ktrain?', batch_size=8)
# top answer snippet extracted from https://arxiv.org/abs/2004.10703:
# "ktrain is a low-code platform for machine learning"
- 2020-11-04
- ktrain v0.24.x is released and now includes built-in support for exporting models to ONNX and TensorFlow Lite. See the example notebook for more information.
ktrain is a lightweight wrapper for the deep learning library TensorFlow Keras (and other libraries) to help build, train, and deploy neural networks and other machine learning models. Inspired by ML framework extensions like fastai and ludwig, ktrain is designed to make deep learning and AI more accessible and easier to apply for both newcomers and experienced practitioners. With only a few lines of code, ktrain allows you to easily and quickly:
-
employ fast, accurate, and easy-to-use pre-canned models for
text
,vision
,graph
, andtabular
data:text
data:- Text Classification: BERT, DistilBERT, NBSVM, fastText, and other models [example notebook]
- Text Regression: BERT, DistilBERT, Embedding-based linear text regression, fastText, and other models [example notebook]
- Sequence Labeling (NER): Bidirectional LSTM with optional CRF layer and various embedding schemes such as pretrained BERT and fasttext word embeddings and character embeddings [example notebook]
- Ready-to-Use NER models for English, Chinese, and Russian with no training required [example notebook]
- Sentence Pair Classification for tasks like paraphrase detection [example notebook]
- Unsupervised Topic Modeling with LDA [example notebook]
- Document Similarity with One-Class Learning: given some documents of interest, find and score new documents that are thematically similar to them using One-Class Text Classification [example notebook]
- Document Recommendation Engines and Semantic Searches: given a text snippet from a sample document, recommend documents that are semantically-related from a larger corpus [example notebook]
- Text Summarization: summarize long documents with a pretrained BART model - no training required [example notebook]
- End-to-End Question-Answering: ask a large text corpus questions and receive exact answers [example notebook]
- Easy-to-Use Built-In Search Engine: perform keyword searches on large collections of documents [example notebook]
- Zero-Shot Learning: classify documents into user-provided topics without training examples [example notebook]
- Language Translation: translate text from one language to another [example notebook]
vision
data:- image classification (e.g., ResNet, Wide ResNet, Inception) [example notebook]
- image regression for predicting numerical targets from photos (e.g., age prediction) [example notebook]
graph
data:- node classification with graph neural networks (GraphSAGE) [example notebook]
- link prediction with graph neural networks (GraphSAGE) [example notebook]
tabular
data:- tabular classification (e.g., Titanic survival prediction) [example notebook]
- tabular regression (e.g., predicting house prices) [example notebook]
-
estimate an optimal learning rate for your model given your data using a Learning Rate Finder
-
utilize learning rate schedules such as the triangular policy, the 1cycle policy, and SGDR to effectively minimize loss and improve generalization
-
build text classifiers for any language (e.g., Arabic Sentiment Analysis with BERT, Chinese Sentiment Analysis with NBSVM)
-
easily train NER models for any language (e.g., Dutch NER )
-
load and preprocess text and image data from a variety of formats
-
inspect data points that were misclassified and provide explanations to help improve your model
-
leverage a simple prediction API for saving and deploying both models and data-preprocessing steps to make predictions on new raw data
Please see the following tutorial notebooks for a guide on how to use ktrain on your projects:
- Tutorial 1: Introduction
- Tutorial 2: Tuning Learning Rates
- Tutorial 3: Image Classification
- Tutorial 4: Text Classification
- Tutorial 5: Learning from Unlabeled Text Data
- Tutorial 6: Text Sequence Tagging for Named Entity Recognition
- Tutorial 7: Graph Node Classification with Graph Neural Networks
- Tutorial 8: Tabular Classification and Regression
- Tutorial A1: Additional tricks, which covers topics such as previewing data augmentation schemes, inspecting intermediate output of Keras models for debugging, setting global weight decay, and use of built-in and custom callbacks.
- Tutorial A2: Explaining Predictions and Misclassifications
- Tutorial A3: Text Classification with Hugging Face Transformers
- Tutorial A4: Using Custom Data Formats and Models: Text Regression with Extra Regressors
Some blog tutorials about ktrain are shown below:
ktrain: A Lightweight Wrapper for Keras to Help Train Neural Networks
Text Classification with Hugging Face Transformers in TensorFlow 2 (Without Tears)
Build an Open-Domain Question-Answering System With BERT in 3 Lines of Code
Finetuning BERT using ktrain for Disaster Tweets Classification by Hamiz Ahmed
Tasks such as text classification and image classification can be accomplished easily with only a few lines of code.
Example: Text Classification of IMDb Movie Reviews Using BERT [see notebook]
import ktrain
from ktrain import text as txt
# load data
(x_train, y_train), (x_test, y_test), preproc = txt.texts_from_folder('data/aclImdb', maxlen=500,
preprocess_mode='bert',
train_test_names=['train', 'test'],
classes=['pos', 'neg'])
# load model
model = txt.text_classifier('bert', (x_train, y_train), preproc=preproc)
# wrap model and data in ktrain.Learner object
learner = ktrain.get_learner(model,
train_data=(x_train, y_train),
val_data=(x_test, y_test),
batch_size=6)
# find good learning rate
learner.lr_find() # briefly simulate training to find good learning rate
learner.lr_plot() # visually identify best learning rate
# train using 1cycle learning rate schedule for 3 epochs
learner.fit_onecycle(2e-5, 3)
Example: Classifying Images of Dogs and Cats Using a Pretrained ResNet50 model [see notebook]
import ktrain
from ktrain import vision as vis
# load data
(train_data, val_data, preproc) = vis.images_from_folder(
datadir='data/dogscats',
data_aug = vis.get_data_aug(horizontal_flip=True),
train_test_names=['train', 'valid'],
target_size=(224,224), color_mode='rgb')
# load model
model = vis.image_classifier('pretrained_resnet50', train_data, val_data, freeze_layers=80)
# wrap model and data in ktrain.Learner object
learner = ktrain.get_learner(model=model, train_data=train_data, val_data=val_data,
workers=8, use_multiprocessing=False, batch_size=64)
# find good learning rate
learner.lr_find() # briefly simulate training to find good learning rate
learner.lr_plot() # visually identify best learning rate
# train using triangular policy with ModelCheckpoint and implicit ReduceLROnPlateau and EarlyStopping
learner.autofit(1e-4, checkpoint_folder='/tmp/saved_weights')
Example: Sequence Labeling for Named Entity Recognition using a randomly initialized Bidirectional LSTM CRF model [see notebook]
import ktrain
from ktrain import text as txt
# load data
(trn, val, preproc) = txt.entities_from_txt('data/ner_dataset.csv',
sentence_column='Sentence #',
word_column='Word',
tag_column='Tag',
data_format='gmb',
use_char=True) # enable character embeddings
# load model
model = txt.sequence_tagger('bilstm-crf', preproc)
# wrap model and data in ktrain.Learner object
learner = ktrain.get_learner(model, train_data=trn, val_data=val)
# conventional training for 1 epoch using a learning rate of 0.001 (Keras default for Adam optmizer)
learner.fit(1e-3, 1)
Example: Node Classification on Cora Citation Graph using a GraphSAGE model [see notbook]
import ktrain
from ktrain import graph as gr
# load data with supervision ratio of 10%
(trn, val, preproc) = gr.graph_nodes_from_csv(
'cora.content', # node attributes/labels
'cora.cites', # edge list
sample_size=20,
holdout_pct=None,
holdout_for_inductive=False,
train_pct=0.1, sep='\t')
# load model
model=gr.graph_node_classifier('graphsage', trn)
# wrap model and data in ktrain.Learner object
learner = ktrain.get_learner(model, train_data=trn, val_data=val, batch_size=64)
# find good learning rate
learner.lr_find(max_epochs=100) # briefly simulate training to find good learning rate
learner.lr_plot() # visually identify best learning rate
# train using triangular policy with ModelCheckpoint and implicit ReduceLROnPlateau and EarlyStopping
learner.autofit(0.01, checkpoint_folder='/tmp/saved_weights')
Example: Text Classification with Hugging Face Transformers on 20 Newsgroups Dataset Using DistilBERT [see notebook]
# load text data
categories = ['alt.atheism', 'soc.religion.christian','comp.graphics', 'sci.med']
from sklearn.datasets import fetch_20newsgroups
train_b = fetch_20newsgroups(subset='train', categories=categories, shuffle=True)
test_b = fetch_20newsgroups(subset='test',categories=categories, shuffle=True)
(x_train, y_train) = (train_b.data, train_b.target)
(x_test, y_test) = (test_b.data, test_b.target)
# build, train, and validate model (Transformer is wrapper around transformers library)
import ktrain
from ktrain import text
MODEL_NAME = 'distilbert-base-uncased'
t = text.Transformer(MODEL_NAME, maxlen=500, class_names=train_b.target_names)
trn = t.preprocess_train(x_train, y_train)
val = t.preprocess_test(x_test, y_test)
model = t.get_classifier()
learner = ktrain.get_learner(model, train_data=trn, val_data=val, batch_size=6)
learner.fit_onecycle(5e-5, 4)
learner.validate(class_names=t.get_classes()) # class_names must be string values
# Output from learner.validate()
# precision recall f1-score support
#
# alt.atheism 0.92 0.93 0.93 319
# comp.graphics 0.97 0.97 0.97 389
# sci.med 0.97 0.95 0.96 396
#soc.religion.christian 0.96 0.96 0.96 398
#
# accuracy 0.96 1502
# macro avg 0.95 0.96 0.95 1502
# weighted avg 0.96 0.96 0.96 1502
Example: Tabular Classification for Titanic Survival Prediction Using an MLP [see notebook]
import ktrain
from ktrain import tabular
import pandas as pd
train_df = pd.read_csv('train.csv', index_col=0)
train_df = train_df.drop(['Name', 'Ticket', 'Cabin'], 1)
trn, val, preproc = tabular.tabular_from_df(train_df, label_columns=['Survived'], random_state=42)
learner = ktrain.get_learner(tabular.tabular_classifier('mlp', trn), train_data=trn, val_data=val)
learner.lr_find(show_plot=True, max_epochs=5) # estimate learning rate
learner.fit_onecycle(5e-3, 10)
# evaluate held-out labeled test set
tst = preproc.preprocess_test(pd.read_csv('heldout.csv', index_col=0))
learner.evaluate(tst, class_names=preproc.get_classes())
Using ktrain on Google Colab? See these Colab examples:
- a simple demo of Multiclass Text Classification with BERT
- a simple demo of Multiclass Text Classification with Hugging Face Transformers
- image classification with Cats vs. Dogs
Additional examples can be found here.
-
Make sure pip is up-to-date with:
pip install -U pip
-
Install TensorFlow 2 if it is not already installed (e.g.,
pip install tensorflow
) -
Install ktrain:
pip install ktrain
The above should be all you need on Linux systems and cloud computing environments like Google Colab and AWS EC2. If you are using ktrain on a Windows computer, you can follow these more detailed instructions that include some extra steps.
Some important things to note about installation:
- If using ktrain with
tensorflow<=2.1
, you must also downgrade the transformers library totransformers==3.1
. - As of v0.21.x, ktrain no longer installs TensorFlow 2 automatically. As indicated above, you should install TensorFlow 2 yourself before installing and using ktrain. On Google Colab, TensorFlow 2 should be already installed. You should be able to use ktrain with any version of TensorFlow 2. Note, however, that there is a bug in TensorFlow 2.2 and 2.3 that affects the Learning-Rate-Finder that will not be fixed until TensorFlow 2.4. The bug causes the learning-rate-finder to complete all epochs even after loss has diverged (i.e., no automatic-stopping).
- If using ktrain on a local machine with a GPU (versus Google Colab, for example), you'll need to install GPU support for TensorFlow 2.
- Since some ktrain dependencies have not yet been migrated to
tf.keras
in TensorFlow 2 (or may have other issues), ktrain is temporarily using forked versions of some libraries. Specifically, ktrain uses forked versions of theeli5
andstellargraph
libraries. If not installed, ktrain will complain when a method or function needing either of these libraries is invoked. To install these forked versions, you can do the following:
pip install git+https://github.com/amaiya/eli5@tfkeras_0_10_1
pip install git+https://github.com/amaiya/stellargraph@no_tf_dep_082
This code was tested on Ubuntu 18.04 LTS using TensorFlow 2.3.1 and Python 3.6.9.
Please cite the following paper when using ktrain:
@article{maiya2020ktrain,
title={ktrain: A Low-Code Library for Augmented Machine Learning},
author={Arun S. Maiya},
year={2020},
eprint={2004.10703},
archivePrefix={arXiv},
primaryClass={cs.LG},
journal={arXiv preprint arXiv:2004.10703},
}
Creator: Arun S. Maiya
Email: arun [at] maiya [dot] net