ssharkov03

"""
The module provides three public classes that implement *Bayesian Invariant Risk Minimization* (BIRM)
training objectives for causal language‑model fine-tuning:

* :class:`EnvironmentEMB` – a lightweight environment embedding that is
  randomly perturbed at every forward pass to approximate the posterior over
  domain‑specific linear classifiers.
* :class:`BIRMCoefficientScheduler` – a helper that dynamically rescales the
  BIRM penalty so that its contribution to the total loss remains in a desired
  ratio to the empirical risk (ERM) term.

ssharkov03

subreddit_name = 'marvel' 
num_comments = 500  # the number of comments you want to get 

import requests
import pandas as pd
from datetime import datetime

# we use this function to convert responses to dataframes
def df_from_response(res):
    # initialize temp dataframe for batch of data in response

ssharkov03

model_mixed = models.alexnet(pretrained=True)

layers_to_unfreeze = 5
for param in model_mixed.features[:-layers_to_unfreeze].parameters():
    param.requires_grad = False
    
num_features = 9216
model_mixed.classifier = nn.Sequential(
    nn.Dropout(p=0.5, inplace=False),
    nn.Linear(num_features, 2, bias=True)

ssharkov03

data_transforms = {
    'train': transforms.Compose([
        transforms.Resize(600),
        transforms.ColorJitter(brightness=.3, hue=.5, contrast=.1, saturation=.1),
        transforms.RandomRotation(degrees=30),
        transforms.RandomPerspective(distortion_scale=.3, p=.7),
        transforms.RandomHorizontalFlip(p=.5),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
        transforms.RandomErasing(p=.3, scale=(0.01, 0.02))