vtjeng

using MIPVerify
using Gurobi

nnparams = get_example_network_params("MNIST.WK17a_linf0.1_authors")
mnist = read_datasets("mnist")

MIPVerify.setloglevel!("info")

MIPVerify.batch_find_untargeted_attack(
    nnparams, 

vtjeng

 ────────────────────────────────────────────────────────────────────────────────────
                                             Time                   Allocations
                                     ──────────────────────   ───────────────────────
          Tot / % measured:                273s / 98.7%           22.0GiB / 98.9%

 Section                     ncalls     time   %tot     avg     alloc   %tot      avg
 ────────────────────────────────────────────────────────────────────────────────────
 integration/                     1     130s  48.1%    130s   9.01GiB  41.4%  9.01GiB
   sequential/                    1     130s  48.0%    130s   9.01GiB  41.4%  9.01GiB
     trained_weights/             1    84.6s  31.4%   84.6s   5.41GiB  24.9%  5.41GiB

vtjeng

 ────────────────────────────────────────────────────────────────────────────────────
                                             Time                   Allocations
                                     ──────────────────────   ───────────────────────
          Tot / % measured:               1795s / 100%            17.9GiB / 98.8%

 Section                     ncalls     time   %tot     avg     alloc   %tot      avg
 ────────────────────────────────────────────────────────────────────────────────────
 integration/                     1    1479s  82.5%   1479s   3.62GiB  20.4%  3.62GiB
   sequential/                    1    1479s  82.5%   1479s   3.62GiB  20.4%  3.62GiB
     generated_weights/           1    1479s  82.5%   1479s   3.62GiB  20.4%  3.62GiB

vtjeng

Utilities

Shell

• zsh, plus:
  • oh_my_zsh and powerlevel9k
  • zsh-autosuggestions plugin: "suggests commands as you type based on history and completions."
• fzf: "general-purpose command-line fuzzy finder"

Source Control

• hub or ghi for working with GitHub over the command line.
• tig for a text-mode interface to git.

vtjeng

This is a guide for local development of a Julia package in Julia 1.0 and above.

tl;dr: We use a new environment to ensure that our regular Julia environment is not affected.

Setting up a new environment

Navigate to ~/.julia/environments. This folder should contain the default environment. (I am currently working in Julia 1.0, so the environments folder contains the folder v1.0).

From the same folder, make a new project by running julia, then entering the Pkg repl by pressing ] from julia.

Upon entering the Pkg REPL, you should see a prompt like this:

vtjeng

# Sources:
# https://www.digitalocean.com/community/tutorials/how-to-harden-openssh-on-ubuntu-18-04
#

PermitRootLogin no

# Limit maximum number of authentication attempts for a particular login session
MaxAuthTries 3

# Prevent certain denial-of-service attacks where multiple authentication sessions are kept open for a prolonged period of time

vtjeng

# Based on https://github.com/LaurentLessard/wordlesolver
# for countmap
using StatsBase

function parse_word_list(filename::String)::Vector{String}
    s = open(filename) do file
        read(file, String)
    end
    s = replace(s, '\"' => "")
    return split(s, ", ")

vtjeng

# A shorter version of zwegner's https://gist.github.com/zwegner/508cc183ab94dd27686a40384783ca4e

# Proof of optimality of 4-move solution to adversarial Wordle
# https://qntm.org/files/wordle/index.html
import collections

with open("wordle-words.txt") as f:
    normal_words = [word.strip() for word in f]

with open("wordle-fake-words.txt") as f:

vtjeng

NAME          
ROWS
 N  OBJ
 L  c1_2
 L  c2_2
 L  c3_2
 L  c4_2
 L  c5_2
 L  c6_2
 L  c7_2