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
Tefla requires Tensorflow(version >=r1.8.0)
pip install tensorflow-gpu
or
pip install tensorflowFor the latest stable version:
pip install teflafor current version installation:
pip install git+https://github.com/openagi/tefla.gitFor Developer / TO Work with source and modifying source code:
git clone https://github.com/openagi/tefla.git
cd tefla
pip install -r requirements.txt
export PYTHONPATH=.Recent deep convolutional models are easy to implement using TEFLA, the state-of-the-art models are implemented using tefla.
- Its as easy as
>>>from tefla.core.layers import conv2d
>>>convolved = conv2d(input, 48, False, None)|-- 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.csv1. [Train v2](https://github.com/openagi/tefla/blob/master/tefla/trainv2.py)
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)- 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(
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 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.