syaffers

class KMeansReducedPalette:
    def __init__(self, num_colors):
        self.num_colors = num_colors
        # Random state for reproducibility.
        self.kmeans = KMeans(num_colors, random_state=0xfee1600d)
        self.source_pixels = None

    def _preprocess(self, image):
        assert image.shape[-1] == 3, 'image must have exactly 3 color channels'
        assert image.dtype == 'uint8', 'image must be in np.uint8 type'

syaffers

import numpy as np
import matplotlib.pyplot as plt

# Load samples.
samples = np.load('data/hg19_HLA_train.npz')['data']

# Probability of randomly deleting genotypes.
noise_prob = 0.05

# Some variants are multiallellic which means that the genotype numbers

syaffers

import io
import subprocess
from sklearn.model_selection import train_test_split
import numpy as np
import tqdm
import vcf


# List of coordinates for the MHC gene.
coordinates = '''

syaffers

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

meta = pd.read_csv('metadata.csv', index_col='ID')

n = 693
data = np.load(meta.loc[n, 'Filename'])
clearing_time = meta.loc[n, 'Clearing Time']

syaffers

def extract_windows_downsampling(array, clearing_time_index, max_time_steps,
                                 sub_window_size, downsampling_ratio):
    start = clearing_time_index + 1 - (sub_window_size - 1) * downsampling_ratio
    
    K_indices = np.arange(0, sub_window_size * downsampling_ratio,
                          step=downsampling_ratio)
    T_indices = np.arange(0, (max_time_steps + 1) * downsampling_ratio,
                          step=downsampling_ratio)
    sub_windows = np.round(
        start + 

syaffers

def vectorized_stride_v1(array, clearing_time_index, max_time, sub_window_size,
                         stride_size):
    start = clearing_time_index + 1 - sub_window_size + 1
    
    sub_windows = (
        start + 
        np.expand_dims(np.arange(sub_window_size), 0) +
        np.expand_dims(np.arange(max_time + 1), 0).T
    )
    

syaffers

def extract_windows_vectorized(array, clearing_time_index, max_time, sub_window_size):
    start = clearing_time_index + 1 - sub_window_size + 1
    
    sub_windows = (
        start +
        # expand_dims are used to convert a 1D array to 2D array.
        np.expand_dims(np.arange(sub_window_size), 0) +
        np.expand_dims(np.arange(max_time + 1), 0).T
    )
    

syaffers

def extract_windows(array, clearing_time_index, max_time, sub_window_size):
    examples = []
    start = clearing_time_index + 1 - sub_window_size + 1
    
    for i in range(max_time+1):
        example = array[start+i:start+sub_window_size+i]
        examples.append(np.expand_dims(example, 0))
    
    return np.vstack(examples)

syaffers

syaffers

import string
from torch.utils.data import DataLoader, random_split
from datasets import TESNamesDataset

data_root = '/home/syafiq/Data/tes-names/'
charset = string.ascii_letters + "-' "
length = 30

dataset = TESNamesDataset(data_root, charset, length)
trainset, valset = random_split(dataset, [15593, 3898])