Define seed for training/finetuning processes and for a reproducible forecast

[dafmdev]

How could a seed be defined for the reproducibility of results?

[ashok-arjun]

Sorry for the delay I didn't monitor Github discussions until now.

I believe you've to add the random seed function before this line.

I recommend seeding everything (random, torch, numpy etc.). I usually use this (but I haven't tested it for your usecase):

import os
import random
import numpy as np
import torch


def set_seed(seed, deterministic=True):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    if torch.cuda.is_available():
        torch.cuda.manual_seed(seed)
        torch.cuda.manual_seed_all(seed)
        torch.backends.cudnn.deterministic = deterministic
        torch.backends.cudnn.benchmark = False
    os.environ["PYTHONHASHSEED"] = str(seed)