williamFalcon / Test Tube

Test Tube

Log, organize and parallelize hyperparameter search for Deep Learning experiments

Docs

View the docs here

Test tube is a python library to track and parallelize hyperparameter search for Deep Learning and ML experiments. It's framework agnostic and built on top of the python argparse API for ease of use.

pip install test_tube

Main test-tube uses

• Parallelize hyperparameter optimization (across multiple gpus or cpus).
• Parallelize hyperparameter optimization across HPC cluster using SLURM.
• Log experiment hyperparameters and experiment data.
  Experiments across models.
• Visualize with tensorboard

Compatible with Python any Python ML library like Tensorflow, Keras, Pytorch, Caffe, Caffe2, Chainer, MXNet, Theano, Scikit-learn

Examples

The Experiment object is a subclass of Pytorch.SummaryWriter.

Log and visualize with Tensorboard

from test-tube import Experiment
import torch

exp = Experiment('/some/path')
exp.tag({'learning_rate': 0.02, 'layers': 4})    

# exp is superclass of SummaryWriter
features = torch.Tensor(100, 784)
writer.add_embedding(features, metadata=label, label_img=images.unsqueeze(1))

# simulate training
for n_iter in range(2000):
    e.log({'testtt': n_iter * np.sin(n_iter)})

# save and close
exp.save()
exp.close()

pip install tensorflow   

tensorboard --logdir /some/path

Run grid search on SLURM GPU cluster

from test_tube.hpc import SlurmCluster

# hyperparameters is a test-tube hyper params object
hyperparams = args.parse()

# init cluster
cluster = SlurmCluster(
    hyperparam_optimizer=hyperparams,
    log_path='/path/to/log/results/to',
    python_cmd='python3'
)

# let the cluster know where to email for a change in job status (ie: complete, fail, etc...)
cluster.notify_job_status(email='[email protected]', on_done=True, on_fail=True)

# set the job options. In this instance, we'll run 20 different models
# each with its own set of hyperparameters giving each one 1 GPU (ie: taking up 20 GPUs)
cluster.per_experiment_nb_gpus = 1
cluster.per_experiment_nb_nodes = 1

# run the models on the cluster
cluster.optimize_parallel_cluster_gpu(train, nb_trials=20, job_name='first_tt_batch', job_display_name='my_batch')   

# we just ran 20 different hyperparameters on 20 GPUs in the HPC cluster!!    

Optimize hyperparameters across GPUs

from test_tube import HyperOptArgumentParser

# subclass of argparse
parser = HyperOptArgumentParser(strategy='random_search')
parser.add_argument('--learning_rate', default=0.002, type=float, help='the learning rate')

# let's enable optimizing over the number of layers in the network
parser.opt_list('--nb_layers', default=2, type=int, tunable=True, options=[2, 4, 8])

# and tune the number of units in each layer
parser.opt_range('--neurons', default=50, type=int, tunable=True, low=100, high=800, nb_samples=10)

# compile (because it's argparse underneath)
hparams = parser.parse_args()

# optimize across 4 gpus
# use 2 gpus together and the other two separately
hparams.optimize_parallel_gpu(MyModel.fit, gpu_ids=['1', '2,3', '0'], nb_trials=192, nb_workers=4)

Or... across CPUs

hparams.optimize_parallel_cpu(MyModel.fit, nb_trials=192, nb_workers=12)

You can also optimize on a log scale to allow better search over magnitudes of hyperparameter values, with a chosen base (disabled by default). Keep in mind that the range you search over must be strictly positive.

from test_tube import HyperOptArgumentParser

# subclass of argparse
parser = HyperOptArgumentParser(strategy='random_search')

# Randomly searches over the (log-transformed) range [100,800).

parser.opt_range('--neurons', default=50, type=int, tunable=True, low=100, high=800, nb_samples=10, log_base=10)


# compile (because it's argparse underneath)
hparams = parser.parse_args()

# run 20 trials of random search over the hyperparams
for hparam_trial in hparams.trials(20):
    train_network(hparam_trial)

Convert your argparse params into searchable params by changing 1 line

import argparse
from test_tube import HyperOptArgumentParser

# these lines are equivalent
parser = argparse.ArgumentParser(description='Process some integers.')
parser = HyperOptArgumentParser(description='Process some integers.', strategy='grid_search')

# do normal argparse stuff
...

Log images inline with metrics

# name must have either jpg, png or jpeg in it
img = np.imread('a.jpg')
exp.log('test_jpg': img, 'val_err': 0.2)

# saves image to ../exp/version/media/test_0.jpg
# csv has file path to that image in that cell

Demos

• Hyperparameter optimization for PyTorch across 20 cluster GPUs
• Hyperparameter optimization across 20 cluster CPUs
• Experiments and hyperparameter optimization for tensorflow across 4 GPUs simultaneously

How to contribute

Feel free to fix bugs and make improvements! 1. Check out the current bugs here or feature requests. 2. To work on a bug or feature, head over to our project page and assign yourself the bug. 3. We'll add contributor names periodically as people improve the library!

Bibtex

To cite the framework use:

@misc{Falcon2017,
  author = {Falcon, W.A.},
  title = {Test Tube},
  year = {2017},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/williamfalcon/test-tube}}
}    

License

In addition to the terms outlined in the license, this software is U.S. Patent Pending.