Counting 3,301 Big Data & Machine Learning Frameworks, Toolsets, and Examples...
Suggestion? Feedback? Tweet @stkim1

Last Commit
Nov. 5, 2018
Nov. 14, 2016

Build Status Build Status PyPI version Build Status


Tefla is built on top of Tensorflow for fast prototyping of deep learning algorithms. It provides higher level access to tensorflow's features. Inerface, Easy to build complex models.

Tefla features:

    - Support for custom optimizers

    . Support for data-sets, data-augmentation
    . Support for text datasets

    . Easy to define complex deep models

    . Single and multi GPU training

    . Various prediction functions including ensembling of models

    . Different metrics for performance measurement

    . Custom losses

    . Learning rate schedules, polynomial, step, validation_loss based

    . Semantic segmentation learning

    . Semi-supervised learning

TensorFlow Installation

Tefla requires Tensorflow(version >=r1.8.0)

pip install tensorflow-gpu
pip install tensorflow

Tefla Installation version

Tefla 1.9.0 released

For the latest stable version:

pip install tefla

for current version installation:

pip install git+

For Developer / TO Work with source and modifying source code:

git clone
cd tefla
pip install -r requirements.txt


Tefla Docs

Tefla Models

Recent deep convolutional models are easy to implement using TEFLA, the state-of-the-art models are implemented using tefla.

  1. Recent Models

Getting Started

  1. Its as easy as
>>>from tefla.core.layers import conv2d
>>>convolved = conv2d(input, 48, False, None)

2a. Data Directory structure for using normal images

|-- Data_Dir
|   |-- training_image_size (eg. training_256, for 256 image size)
|   |-- validation_image_size (eg. validation_256, for 256 image size)
|   |-- training_labels.csv
|   |-- validation_labels.csv

2b. TFRecords support available using tefla/dataset class

1. [Train v2](

Run training:

python tefla/ --model models/ --training_cnf models/ --data_dir /path/to/data/dir (as per instructions 2.a)
  1. Mnist example gives a overview about Tefla usages
image_size =(32, 32)
crop_size = (28, 28)
def model(is_training, reuse):
    common_args = common_layer_args(is_training, reuse)
    conv_args = make_args(batch_norm=True, activation=prelu, **common_args)
    fc_args = make_args(activation=prelu, **common_args)
    logit_args = make_args(activation=None, **common_args)

    x = input((None, height, width, 1), **common_args)
    x = conv2d(x, 32, name='conv1_1', **conv_args)
    x = conv2d(x, 32, name='conv1_2', **conv_args)
    x = max_pool(x, name='pool1', **common_args)
    x = dropout(x, drop_p=0.25, name='dropout1', **common_args)
    x = fully_connected(x, n_output=128, name='fc1', **fc_args)
    x = dropout(x, drop_p=0.5, name='dropout2', **common_args)
    logits = fully_connected(x, n_output=10, name="logits", **logit_args)
    predictions = softmax(logits, name='predictions', **common_args)

    return end_points(is_training)

training_cnf = {
    'classification': True,
    'validation_scores': [('validation accuracy', util.accuracy_wrapper), ('validation kappa', util.kappa_wrapper)],
    'num_epochs': 50,
    'lr_policy': StepDecayPolicy(
            0: 0.01,
            30: 0.001,
util.init_logging('train.log', file_log_level=logging.INFO, console_log_level=logging.INFO)

trainer = SupervisedTrainer(model, training_cnf, classification=training_cnf['classification']), weights_from=None, start_epoch=1, verbose=1, summary_every=10)


Welcome to the fourth release of Tefla, if you find any bug, please report it in the GitHub issues section.

Improvements and requests for new features are more than welcome! Do not hesitate to twist and tweak Tefla, and send pull-requests.


MIT License

Latest Releases
 Sep. 17 2018
 Sep. 14 2018
 Aug. 31 2018
 Aug. 26 2018
 Aug. 26 2018