N-McA

import collections
import dataclasses
import itertools
import typing as tp
from functools import partial

import numpy as np
from scipy.optimize import minimize

DEFAULT_CI_SIZE = 0.95

N-McA

import numpy as np

import theano.tensor as T
import pymc3 as pm

from utils import add_params_property

import ranking
from ranking import tennis_data, MPTrueSkill1V1NoDrawRanker

N-McA

#!/usr/bin/env python3

import subprocess
import shlex
import os
from pathlib import Path


def chdir_to_script_location():
    abspath = os.path.abspath(__file__)

N-McA

#!/usr/bin/env python3

import subprocess
import shlex
import os
from pathlib import Path


def chdir_to_script_location():
    abspath = os.path.abspath(__file__)

N-McA

#!/usr/bin/env python3

import subprocess
import shlex
import os
from pathlib import Path


def chdir_to_script_location():
    abspath = os.path.abspath(__file__)

N-McA

function isString(s) {
  return (typeof s === 'string' || s instanceof String)
}

export function toBaseUnit(value, decimals, BN) {
  if (!isString(value)) {
    throw new Error('Pass strings to prevent floating point precision issues.')
  }
  const ten = new BN(10);
  const base = ten.pow(new BN(decimals));

N-McA

import keras.backend as kb
from keras.layers import Layer

def _kb_linspace(num):
    num = kb.cast(num, kb.floatx())
    return kb.arange(0, num, dtype=kb.floatx()) / (num - 1)

def _kb_grid_coords(width, height):
    w, h = width, height
        

N-McA

def tf_pca(x):
    '''
        Compute PCA on the bottom two dimensions of x,
        eg assuming dims = [..., observations, features]
    '''
    # Center
    x -= tf.reduce_mean(x, -2, keepdims=True)

    # Currently, the GPU implementation of SVD is awful.
    # It is slower than moving data back to CPU to SVD there

N-McA

'''
Compatible with Keras, faster than reading from files (no stats).
It's only designed to work if all your images are vaguely similar sizes/
when encoded as JPGS, so if you have:
    white noise or other hard-to-encode stuff
    radically varying image sizes
    ...
    (probably other failure modes)
Then this is foolhardy.

N-McA

header = r'''
\begin{tikzpicture}[node distance = 2mm, auto]
%% Auto Generated
'''

raw_b = r'''
\node [block, below= of glove] (conv1) {
    \begin{tabular}{cc}
        Conv1D & Input:  $n$x100 \\
        64x5  Dilation 1 & Output: $n$x64  \\