| # Fast block jackknife for linear regression (the LDSC trick): | |
| # compute block-wise XtX and Xty once, then get all 200 leave-one-block-out fits | |
| # by subtracting 1 block at a time and re-solving. | |
| block_jackknife_lm <- function(X, y, n_blocks = 200L, separators = NULL) { | |
| X <- as.matrix(X) | |
| y <- matrix(as.numeric(y), ncol = 1) | |
| n <- nrow(X); p <- ncol(X) | |
| stopifnot(nrow(y) == n, ncol(y) == 1, p <= n) |
| import os | |
| import numpy as np | |
| import sidechainnet as scn | |
| from denoising_diffusion_pytorch import Unet, GaussianDiffusion | |
| import torch | |
| from torch.utils.data import Dataset, DataLoader | |
| import torchvision.transforms as T | |
| from PIL import Image | |
| from pathlib import Path | |
| from tqdm import tqdm |
| # Full training script for protein contact map diffusion model | |
| # Using LucidRain's denoising-diffusion-pytorch (grayscale input) | |
| from denoising_diffusion_pytorch import Unet, GaussianDiffusion, Trainer | |
| import matplotlib.pyplot as plt | |
| import numpy as np | |
| model = Unet( |
| # t-test vs wilcox vs ordinal | |
| library(tidyverse) | |
| library(multidplyr) # parallelize | |
| library(rms) # ordinal regression | |
| sample_size = 500 # N | |
| # genarate paired sets and calculate p-values with different techniques |
| ## The problem with (personal) non-ordinality: | |
| n <- 30000 # 10k fictional people | |
| # a pair of exposures, no measurmennt issues: | |
| xa <- rnorm(n) | |
| xb <- rnorm(n) | |
| # Personal threshold 1 - 4 for each person, | |
| # reasonable scale design I think by which I mean the bins fill up sort of "normal" like this matters a lot!! |
| require(MASS) | |
| # fixed a2 and e2 for the entire script: | |
| a <- .87 # additive genetic variance | |
| e <- .13 # environmental variance | |
| # make ZM covariance, i.e. the cov is a, the var = 1 | |
| sigma_mz <- matrix(c(1,a,a,1),2,2) | |
| # sample size (is big becuase rare traits) |
| import torch | |
| from transformers import GPT2Tokenizer, GPT2LMHeadModel | |
| # Load the tokenizer and model | |
| tokenizer = GPT2Tokenizer.from_pretrained('gpt2') | |
| model = GPT2LMHeadModel.from_pretrained('./results') | |
| # Set the device to GPU if available, otherwise use CPU | |
| device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') | |
| model.to(device) |
| import torch | |
| from torch.utils.data import Dataset, DataLoader | |
| from transformers import GPT2Tokenizer, GPT2LMHeadModel, TextDataset, DataCollatorForLanguageModeling, Trainer, TrainingArguments | |
| # Set the path to the text file to fine-tune on | |
| path_to_file = "path/to/text/file.txt" | |
| # Load the tokenizer and model | |
| tokenizer = GPT2Tokenizer.from_pretrained('gpt2') | |
| model = GPT2LMHeadModel.from_pretrained('gpt2') |
Speaker 0:
You wrote a piece a follow-up piece to your oral history titled, there is no replacement for black Twitter. I think back in November, What do you think we lose if we lose black Twitter? Tell
Speaker 1:
me not to meet your Mac, but we lose everything. I'm John Favreau. Welcome to offline.
Speaker 0: