torchminer 0.4.14

Run Torch With A Simple Miner

Published on pypi

Packaged Using Poetry

Description

TorchMiner is designed to automatic process the training ,evaluating and testing process for PyTorch DeepLearning,with a simple API.

You can access all Functions of MineTorch simply use Miner.

Quick Start

import TorchMiner
from TorchMiner import Miner
from TorchMiner.plugins.Logger.Jupyter import JupyterLogger, JupyterTqdm
from TorchMiner.plugins.Metrics import MultiClassesClassificationMetric
from TorchMiner.plugins.Recorder import TensorboardDrawer

miner = Miner(
    alchemy_directory='/the/route/to/log', 
    train_dataloader=train_dataloader, 
    val_dataloader=val_dataloader,  

    model=model, 
    loss_func=MSELoss,  
    optimizer=optimizer,  
    # or, by passing a function to optimizer, TorchMiner can auto cuda the params of optimizer
    # optimizer=lambda x: optim.SGD(x.parameters(), lr=0.01)，
    experiment="the-name-of-experiment",  # Subdistribution in the experimental directory
    resume=True,  # Whether to automatically load the previous model
    eval_epoch=1,  # How many rounds are evaluated
    persist_epoch=2,  # How many rounds are saved once a checkpoint
    accumulated_iter=1,  # How many times iterates the parameter update after accumulation
    in_notebook=True,
    amp=True,  # Whether to use amp
    plugins=[
        # Use the plugins to extend the function of miner
        JupyterLogger(),
        JupyterTqdm(),
        # or, you can use the below one to auto enable the above two
        # *JupyterEnvironmentAutoEnable(),
        # The two above plugins are designed to get better output in Jupyter Enviroment
        MultiClassesClassificationMetric(),
        # This Plugin can automaticly calculate Accuracy, kappa score and Confusion Matrix in Classification problems.
        TensorboardDrawer(input_to_model),
        # This Plugin can record the informations generate by training process or by other plugins in Tensorboard.
    ],
)

# And then, trigger the training process by
miner.train()