/Zeras

A package of deep-learning utilities

Primary LanguagePythonMIT LicenseMIT

Zeras

A package of deep-learning utilities.

Description

Common to tf and torch:

from Zeras.data_tools import DataBatcher, DataParallelism
from Zeras.data_tools import get_files_with_ext
from Zeras.data_tools import save_data_to_pkl, load_data_from_pkl
from Zeras.data_paral_batcher import DataParalBatcher

from Zeras.vocab import Vocab

from Zeras.settings_baseboard import SettingsBaseboard

Specific to tfv1:

from Zeras.model_baseboard import ModelBaseboard
from Zeras.model_baseboard import initialize_from_ckpt
from Zeras.model_baseboard import get_all_variables
from Zeras.model_baseboard import remove_from_trainable_variables

from Zeras.optim import constant_lr
from Zeras.optim import linear_warmup_and_exp_decayed_lr
from Zeras.optim import linear_warmup_and_polynomial_decayed_lr
from Zeras.optim import adam_optimizer
from Zeras.optim import adam_wd_optimizer
from Zeras.optim import AdamWeightDecayOptimizer

import Zeras.zoo_nn as zrs_nn
import Zeras.zoo_layers as zrs_layers

Installation

From PyPI distribution system:

pip install Zeras

Or from the code:

git clone https://github.com/Li-Ming-Fan/Zeras
cd Zeras
make clean install

Or just incorporate the directory Zeras/ to your project.

Usage

For usage of this package, please refer to the repositories:

https://github.com/Li-Ming-Fan/text_classification

https://github.com/Li-Ming-Fan/transformer-tensorflow

https://github.com/Li-Ming-Fan/pointer-generator-refactored


Acknowledgement

The name "Zeras" is a variant from Keras. We use "Z" because it makes the direcory Zeras/ always display in the last (or first) and do not intersect other directories.