zenkai 0.0.0

Your package description

Zenkai

Zenkai is a framework to extend the internal mechanics and training mechanics for deep learning machines. It is largely built on top of PyTorch.

Installation

pip install zenkai

Usage

Zenkai's primary feature is the "LearningMachine" which aims to make defining learning machines flexible. The design is similar to Torch, in that there is a forward method, a parameter update method similar to accGradParameters(), and a backpropagation method similar to updateGradInputs(). So the primary usage will be to implement them.

Here is a (non-working) example

class MyLearner(zenkai.LearningMachine):
    """A LearningMachine couples the learning mechanics for the machine with its internal mechanics."""

    def __init__(
        self, module: nn.Module, step_theta: zenkai.StepTheta, 
        step_x: StepX, loss: zenkai.Loss
    ):
        super().__init__()
        self.module = module
        # step_theta is used to update the parameters of the
        # module
        self.step_theta = step_theta
        # step_x is used to update the inputs to the module
        self.step_x = step_x
        self.loss = loss

    def assess_y(
        self, x: zenkai.IO, t: zenkai.IO, reduction_override: str=None
    ) -> zenkai.AssessmentDict:
        # assess_y evaluates the learning machine
        return self.loss.assess_dict(x[0], t[0], reduction_override)

    def step(
        self, conn: zenkai.Conn, state: zenkai.State, from_: zenkai.IO
    ) -> zenkai.Conn:
        # step is analogous to accGradParameters in Torch but more general
        # Conn (Connection) is an object that contains the inputs and
        # outputs for a connection of two machines
        return self.step(conn, state, from_)

    def step_x(
        self, conn: zenkai.Conn, state: zenkai.State
    ) -> zenkai.Conn:
        # step_x is analogous to updateGradInputs in Torch except
        # it calculates "new targets" for the incoming layer
        return self.step_x(conn, state)

    def forward(self, x: zenkai.IO) -> zenkai.IO:
        return zenkai.IO(self.module[x[0]])


my_learner = MyLearner(...)
assessment = my_learner.learn(x, t)

Zenkai consists of stacking learning machines in order

Contributing

To contribute to the project

1. Fork the project
2. Create your feature branch
3. Commit your changes
4. Push to the branch
5. Open a pull request

License

This project is licensed under the MIT License - see the LICENSE.md file for details.

Citing this Software

If you use this software in your research, we request you cite it. We have provided a CITATION.cff file in the root of the repository. Here is an example of how you might use it in BibTeX: