vadimkantorov

import sys
import re
import mailbox

emails = set()
for message in mailbox.mbox(sys.argv[1]).itervalues():
    for k, v in message.items():
        if k.lower() in ['from', 'to', 'cc']:
            emails |= set(re.split(r''' |<|>|"|,|\t''', v))

vadimkantorov

# Use Chrome Session Buddy plugins to export and import tab URL lists
# alias splittabs='python3 <(curl -s https://gist.githubusercontent.com/vadimkantorov/ab9acf28bdb0b5f3423973106a4d13a5/raw)'
# splittabs session_buddy.txt --noperc

import sys

input_file_path, noperc = sys.argv[1], (sys.argv + [None])[2] == '--noperc'
lines = set(filter(bool, open(input_file_path)))

def flt(name, lines, contains = []):

vadimkantorov

in:inbox is:unread category:primary

vadimkantorov

# Does not work because of https://github.com/twintproject/twint/issues/1011
# # Prerequisite: https://github.com/twintproject/twint
# # Usage: dump_twitter_followed_bio vadimkantorov > followed.txt
# # Takes 1h for 2500 followed accouns
# # alias dump_twitter_followed_bio='twint --following --user-full -u'

# Works:
# Save in Chrome HAR session of scrolling the Followers / Following timeline

import argparse

vadimkantorov

set -e

# for f in *.m4a; do ffmpeg -i "$f" -af "silenceremove=stop_periods=-1:stop_duration=0.2:stop_threshold=-40dB:window=1" -y "nosilence/$f"; done

for f in $1/*.wav; do
    fname=$(basename $f)
    segdir=segments.$fname
    rm -rf $segdir
    mkdir $segdir
    for channel in 0 1; do

vadimkantorov

# https://superuser.com/questions/650291/how-to-get-video-duration-in-seconds
# Usage: duration myfile.wav

alias duration="ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1"

# mkdir -p sample; find segments.* -name '*.wav' -type f | while read filename; do echo "$(duration $filename) $filename"; done | grep "^5\." | while read duration_ filename_; do cp $filename_ sample; done

vadimkantorov

import torch

class ReLUDropout(torch.nn.Dropout):
    def forward(self, input):
        return relu_dropout(input, p = self.p, training = self.training, inplace = self.inplace)

def relu_dropout(x, p = 0, inplace = False, training = False):
    if not training or p == 0:
        return x.clamp_(min = 0) if inplace else x.clamp(min = 0)

vadimkantorov

# Usage: python3 split_audio_by_silence.py -i input_audio.m4a -o segments
#  will save segments in mp3 format into the segments directory
#  based on https://github.com/mozilla/DeepSpeech/tree/master/examples/vad_transcriber
# Dependencies: webrtcvad

import os
import argparse
import collections
import subprocess
import webrtcvad

vadimkantorov

import torch

# ported from https://github.com/NVIDIA/OpenSeq2Seq/blob/master/open_seq2seq/optimizers/lr_policies.py and https://www.tensorflow.org/api_docs/python/tf/train/polynomial_decay
class PolynomialDecayLR(torch.optim.lr_scheduler._LRScheduler):
    def __init__(self, optimizer, decay_steps, power = 1.0, begin_decay_at = 0, end_lr = 0.0, warmup_steps = 0, last_epoch = -1):
        self.decay_steps = decay_steps
        self.power = power
        self.begin_decay_at = begin_decay_at
        self.end_lr = end_lr
        self.warmup_steps = warmup_steps

vadimkantorov

# ported from https://github.com/NVIDIA/OpenSeq2Seq/blob/master/open_seq2seq/optimizers/novograd.py
# paper: https://arxiv.org/abs/1905.11286
# a recent NVidia's implementation in PyTorch: https://github.com/NVIDIA/DeepLearningExamples/blob/master/PyTorch/SpeechRecognition/Jasper/optimizers.py

import torch

class NovoGrad(torch.optim.Optimizer):
    def __init__(self, params, lr=1.0, betas = (0.95, 0.98), eps=1e-8, weight_decay=0.0, dampening=False):
        defaults = dict(lr=lr, betas=betas, eps=eps, weight_decay=weight_decay, dampening=dampening)
        super(NovoGrad, self).__init__(params, defaults)