Skip to main content

No project description provided

Project description

pyklopp Maintenance Python 3.6 Python 3.7 Python 3.6 Tests

Tired of logging all hyperparameter configurations of your model prototyping to disk?

Pyklopp is a tool to initialize, train and evaluate pytorch models (currently for supervised problems). It persists all relevant hyperparameters, timings and model configurations. Your prototyping is reduced to defining your model, the dataset and your desired parameters.

Important note: we are undergoing an architectural change from writing config json files to writing meta data files given a jsonschema. So to keep your experiments reproducible and program against a current design of pyklopp, reference the exact pyklopp version in your experiment. E.g. for your environment.yml:

dependencies:
- pip:
  - pyklopp==0.3.0

Workflow sketch for developing a model and running it with pyklopp.

Installation

You can install a version from PyPi with: pip install pyklopp.

To install the latest development build, you can clone the repository and invoke poetry build (having poetry installed). Then you can use the built package and install it with pip in your current environment by pip install dist/xxx.whl.

Defining model & dataset

Used imports:

import pypaddle.sparse
import pypaddle.util
import torch.nn as nn
import torch.nn.functional as F

Specify your model in a plain python file, e.g.:

# my_model.py

# Your model can be any pytorch module
# Make sure to not define it locally (e.g. within the get_model()-function)
class LeNet(nn.Module):
    def __init__(self, output_size):
        super(LeNet, self).__init__()
        self.conv1 = nn.Conv2d(3, 6, 5)
        self.conv2 = nn.Conv2d(6, 16, 5)
        self.fc1 = nn.Linear(16 * 5 * 5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, output_size)

    def forward(self, x):
        out = F.relu(self.conv1(x))
        out = F.max_pool2d(out, 2)
        out = F.relu(self.conv2(out))
        out = F.max_pool2d(out, 2)
        out = out.view(out.size(0), -1)
        out = F.relu(self.fc1(out))
        out = F.relu(self.fc2(out))
        out = self.fc3(out)
        return out


# This is your model-instantiation function
# It receives an assembled configuration keyword argument list and should return an instance of your model
def get_model(**kwargs):
    output_size = int(kwargs['output_size'])

    return LeNet(output_size)

Invoke pyklopp to initialize it: pyklopp init my_model.get_model --save='test/model.pth' --config='{"output_size": 10}' Train it on cifar10:

  • pyklopp train test/model.pth cifar10.py --save='test/trained.pth' --config='{"batch_size": 100}'
  • pyklopp train test/model.pth torchvision.datasets.cifar.CIFAR10 --save 'test/trained.pth' --config='{"dataset_root": "/media/data/set/cifar10"}'

Examples

# Initializing & Saving: mymodel.py
pyklopp init foo --save='mymodel1/model.pth'
pyklopp init foo --config='{"python_seed_initial": 100}' --save='mymodel2/model.pth'

# Training
pyklopp train path/to/mymodel.pth mnist
pyklopp train path/to/mymodel.pth mnist --config='{"batch_size": 100, "learning_rate": 0.01}'
# foo.py - Your model initialization function

def init(**kwargs):
    input_size = kwargs['input_size']
    output_size = kwargs['output_size']

    return pypaddle.sparse.MaskedDeepFFN(input_size, output_size, [100, 100])
# mnist.py - Your dataset loading functions

def train_loader(**kwargs):
    batch_size = kwargs['batch_size']

    mnist_train_loader, mnist_test_loader, _, selected_root = pypaddle.util.get_mnist_loaders(batch_size, '/media/data/set/mnist')
    return mnist_train_loader


def test_loader(**kwargs):
    batch_size = kwargs['batch_size']

    mnist_train_loader, mnist_test_loader, _, selected_root = pypaddle.util.get_mnist_loaders(batch_size, '/media/data/set/mnist')
    return mnist_test_loader

Development

  • Create wheel files in dist/: poetry build
  • Install wheel in current environment with pip: pip install path/to/pyklopp/dist/pyklopp-0.1.0-py3-none-any.whl

Running CI image locally

Install latest gitlab-runner (version 12.3 or up):

# For Debian/Ubuntu/Mint
curl -L https://packages.gitlab.com/install/repositories/runner/gitlab-runner/script.deb.sh | sudo bash

# For RHEL/CentOS/Fedora
curl -L https://packages.gitlab.com/install/repositories/runner/gitlab-runner/script.rpm.sh | sudo bash

apt-get update
apt-get install gitlab-runner

$ gitlab-runner -v
Version:      12.3.0

Execute job tests: gitlab-runner exec docker test-python3.6

Running github action locally

Install https://github.com/nektos/act. Run act

Running pre-commit checks locally

poetry run pre-commit run --all-files

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for pyklopp, version 0.3.2
Filename, size File type Python version Upload date Hashes
Filename, size pyklopp-0.3.2-py3-none-any.whl (25.9 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size pyklopp-0.3.2.tar.gz (18.8 kB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page