hideandseek 0.1.1.12

library for deep learning and privacy preserving deep learning

hideandseek

privacy preserving deep learning library.

Why use hideandseek?

• Easy training & saving deep learning models along with other modules (ex: preprocessing modules) required in inference

• Run multiple deep learning experiments in parallel on multiples GPUs (powered by hydra, and python multiprocessing)

• Design and analyze experiments scientifically by modifying variables (powered by hydra)

• Modularized machine learning pipeline allows using the same script for all types of experiments

• The same training code can be run in privacy preserving setting by minimal modifications

Currently integrating from experiment codes. (30.10.2021.)

import torch
from omegaconf import OmegaConf
import hideandseek as hs

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
cfg = OmegaConf.load('config.yaml') # omegaconf.OmegaConf.DictConfig object
model = DNN() # torch.nn.Module object
train_dataset = dataset # torch.utils.data.Dataset object
kwargs = {
  'model': model,
  'dataset': train_dataset,
  'cfg_train': cfg,
  'criterion': criterion,
}
node = hs.Node(**kwargs)

node.model.to(device)
node.step(local_T=20, horizon='epoch') # trains for 20 epochs
# node.step(local_T=1000, horizon='step') # trains for 1000 steps
node.model.cpu()

node.save()

test_results = hs.eval.test(node)
scores = hs.eval.scores(test_results)

To do

• Migrate modules from experiment codes
• Draw figures to explain hideandseek
• GUI for generating experiment scripts when conducting variable sweeps