openagi / tefla Goto Github PK

Tefla is built on top of Tensorflow for fast prototyping of deep learning algorithms. It provides high level access to the features of tensorflow. An interface to easily build complex models.

Tefla has several advantages over Keras or any other library. You may always ask why Tensorflow instead of PyTorch, Caffe, MXNet, and others? Its an effort from an Indian AI company and AI developers to make AI accessible to all.

1. Tefla allows you to design complex models at ease.

2. Tefla guarantees scalability, portability and reproducibility.

3. Tefla's models repo provides state-of-the-art deep learning methods implementation for free.

4. Tefla makes the deep model deployment very easy; it has inbuilt API to serve the trained model.

5. Tefla comes with API for segmentation learning, semi-supervised learning, and generative model training.

6. Tefla has faster input pipeline.

7. Tefla enables faster model training with Multi-GPU support.

8. Tefla also comes with language modeling tools, encoder, decoder framework support.

9. Tefla also supports reinforcement learning modeling and training.

Tefla features:

    . Supports custom optimizers

    . Supports data-sets, data-augmentation, and others
   
    . Supports 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

Before you install Tefla you need to install Tensorflow version r1.8.0 or later.

pip install tensorflow-gpu
or 
pip install tensorflow

The latest release of Tefla is version 1.9.0.

• To install the latest stable version:

pip install tefla

• To install the current version:

pip install git+https://github.com/openagi/tefla.git

• To develop or work with source and modifying source code:

git clone https://github.com/openagi/tefla.git
cd tefla
pip install -r requirements.txt
export PYTHONPATH=.

For more information about installing, configuring, and managing see, Tefla Docs

Recent deep convolutional models are easy to implement using Tefla. For more information about the latest state-of-the-art models that are implemented using tefla, see Recent Models

1. Import the layers

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

2. Create the data directory

• 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

• TFRecords support available using tefla/dataset class and Train v2

3. Run the training

python tefla/train.py --model models/alexnet.py --training_cnf models/multiclass_cnf.py --data_dir /path/to/data/dir (as per instructions 2.a)

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(
        schedule={
            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'])
trainer.fit(data_set, 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 the issue on GitHub. Improvements and requests for new features are more than welcome. Do not hesitate to twist and tweak Tefla, and send pull-requests.

MIT License