I work on many different projects and I use some ActivityWatch for automatic time tracking. AW has a feature where you can use regexes to match window titles to assign them automatically to different projects. But, unfortunately, Jupyter Notebook have a "{filename} - Jupyter Notebook" title in the browser, so it's hard to match on them. I'd rather not name all the notebook with a prefix of the project, so I went looking for a different solution.

I quickly found a way to customize the title shown in the browser, but not in a dynamic way (aka dependent on the folder I'm in).

To get dynamic custom titles, I had to first update in the config file used by Jupyter (~/.jupyter/jupyter_notebook_config.py) the following two values:

c.NotebookApp.extra_template_paths = ["~/.jupyter/templates/"]

import os
from pathlib import Path

cur_dir = os.getcwd()

projects = {
        'folder_name1': 'Project 1',
        'folder_name2': "Project 2",
}

cur_env = Path(cur_dir).name
for p in projects:
    if p in cur_dir:
        cur_env = projects[p]
    
c.NotebookApp.jinja_template_vars = {"env_name": cur_env}

The first line add a new path where Jupyter looks for HTML templates. Then I get the name of the current working directory. I look for some folder names and if they match, I get the project name, otherwise I'll use as project name the name of the current working directory. On the last line I inject the env_name variable into every Jinja2 template used by Jupyter.

Then I copied into ~/.jupyter/templates/page.html the template file from the notebook project and at the end of the last script block, I added the following:

{% if env_name %}
window.title += " - {{ env_name }}";
document.__defineSetter__('title', function(val) {
    document.querySelector('title').childNodes[0].nodeValue = val + " - {{ env_name }}";
});
{% endif %}

First I check if env_name is set. If it is, I add Javascript code which will add the value of it to the window title, and also will update the window title whenever it changes (such as when you rename the file).

This is a bit hackish and when the notebook templates update, I should update my own copy as well. Luckily, it doesn't change too often, there being only 5 commits since 2020.

I often need to set up a VM with a server to run some ML models. While I could package things in Docker and deploy them like that, whenever possible, I like to do things more directly, using systemd services.

To create a systemd service, create a file called /etc/systemd/system/my_service.service. This will create a service that will be started by the root user, but it can be run under another user.

There are three necessary components of a systemd service: unit, service and install.

[Unit]
Description=Service to run my fancy API
After=network.target
StartLimitIntervalSec=100
StartLimitBurst=5

[Service]

Restart=always
RestartSec=10
User=ubuntu
Group=www-data
WorkingDirectory=/home/ubuntu/app
Environment="PATH=/home/ubuntu/.venv/bin"
Environment="API_KEY=123456"
ExecStart=/home/ubuntu/app/.venv/bin/uvicorn main:app --my_params
StandardOutput=append:/var/log/app/stdout.log
StandardError=append:/var/log/app/std_error.log

[Install]
WantedBy=multi-user.target

First, we have to describe the service. Then we tell it when to start: after the network is up and running. We tell it to stop restarting if it crashes 5 times in 100 seconds.

Then we describe the actual service. We tell it to always restart (keeping the above mentioned limits), waiting 10 seconds between restart. We can specify a user, group, working directory and environment variables in which to run the program.

With ExecStart we can specify the program to run. If it's a simple FastAPI server, I usually write out the full path here. But, if it's something more complicated, where I might have to play around with the command more often, I create a bash file and start the service using that: ExecStart=bash /home/ubuntu/start_service.sh.

Then we redirect the standard output and the error stream to files. The last bit I have no idea what it is, but it's needed.

To start the service you have to first run sudo systemctl daemon-reload. You will have to run this every time you change the service definition file. Then you can start the service with sudo systemctl start my_service (the name of the file, without .service).

To enable starting it automatically on startup, you can run sudo systemctl enable my_service.

To see the status of the service: sudo systemctl status my_service. If it's crashing, you can look at journalctl -xe or you can check the log files you configured.

I recommend always writing out full paths, especially for the Python interpreter, to avoid any ambiguities as to what will run your program or what exactly will be loaded.

And now you know how to create a simple systemd service!

When running long running code in Jupyter, I want to get notified when it finished so that I can get back to it. There is an extension to do that with browser notifications, but there are times when I leave the computer while waiting for an ML training to finish.

For long running CLI commands there is the ntfy, a command line tool that allows you to send notifications through a lot of channels.

So I hacked the two together to get some code that automatically messages me on Telegram when a cell finished more than 60 seconds after it started. This extension is registered automatically on the startup of any IPython and Jupyter notebook (even if they are installed in random virtual environments). Why Telegram? Because I already have it installed and it seemed like the easiest integration to set up.

The code has to be placed in ~\.ipython\profile_default\startup\01notification.py. You can place multiple files in this folder and they are loaded in lexicographic order, so you should prepend a number if you care about order. First, a couple of magic imports:

import time
import subprocess

from IPython.core.getipython import get_ipython
from IPython.core.magic_arguments import argument, magic_arguments, parse_argstring
from IPython.core.magic import (register_line_magic)

To send the notification using ntfy, I'm simply calling the program using subprocess. The path resolution used by subprocess is not clear to me, so I had to use the full path to the executable.

def display_notification(message):
    subprocess.run([r"C:\Users\Roland\AppData\Local\Programs\Python\Python310\Scripts\ntfy.exe", "-b", "telegram", "send", message]) 

Then we define some variables, one for the threshold for notification and the other one for remembering the start of the execution:

autonotify_after = 60
run_start_time = None

And now we define the magic function (that's what these things are called in IPython). It has two arguments, one to override the duration of the threshold for notifications and the other for the default message. I copy pasted the decorators as you see them. After parsing the arguments (which come as a string), we register two event handlers: one to run before a cell is executed and one after a cell is executed.

@magic_arguments()
@argument(
    "-a", "--after", default=None,
    help="Send notification if cell execution is longer than x seconds"
)
@argument(
    "-m",
    "--message",
    default="Cell Execution Has Finished!",
    help="Custom notification message"
)
@register_line_magic
def autonotify(line):
    # Record options
    args = parse_argstring(autonotify, line)
    message = args.message.lstrip("\'\"").rstrip("\'\"")
    if args.after:
        global autonotify_after
        autonotify_after = args.after
    ### Register events
    ip = get_ipython()

    # Register new events
    ip.events.register('pre_run_cell', pre_run_cell)
    ip.events.register('post_run_cell', lambda: post_run_cell(message))

The handler to run before a cell is simple: we just record the start time of the run.

def pre_run_cell():
    global run_start_time
    run_start_time = time.time()

The second handler is slightly more complex. We look at the output of the last cell and append it to the message if it's not an "empty" value. We check how long has elapsed to know whether to show a notification or not:

def post_run_cell(message):
    # Set last output as notification message
    last_output = get_ipython().user_global_ns['_']
    # Don't use output if it's None or empty (but still allow False, 0, etc.)
    try:
        if last_output is not None and len(str(last_output)):
            message = message + "\n" + str(last_output)
    except ValueError:
        pass # can't convert to string. Use default message
    
    # Check autonotify options and perform checks
    if not check_after(): 
        return
    display_notification(message)


def check_after():
    # Check if the time elapsed is over the specified time.
    now, start = time.time(), run_start_time
    return autonotify_after >= 0 and start and (now - start) >= autonotify_after

The last piece of magic is to run this function. The other blog post I was inspired by said you should delete the function for things to work properly, so I did that as well:

ipython = get_ipython()
ipython.magic('autonotify')
del autonotify

And voila, now you will get Telegram notifications automatically when your model  finishes training! Setting up a Telegram bot is left as an exercise for the reader.

Yesterday my friend Marius Corici called me and asked me if I wanted to be interviewed (again) by ProTV, this time about deep fakes. This time it was a Zoom interview, and they asked me not to blur my background (ugh). And I had my second 15 seconds of fame last night (the actual interview was 5 minutes long, but they cut out most of it).

tl;dr: Deep fakes are already here, they are getting easier and easier to make and it's a cat and mouse game to detect them.

Source: ProTV

As I work from home 90% of the time, I run into a small issue during meetings: I sometimes speak too loudly. Before my daughter Gloria arrived, this was something that annoyed my wife and others in the house, but now, when Gloria is sleeping, this is not just an annoyance, it's a BIG problem, because nobody wants to wake up a toddler.

While I do have monitoring that alerts me via Signal that I'm speaking too loud (my wife), I wanted to write a program to do that all the time, in the spirit of using programming to make my life nicer.

So I started to look for a Python library that can give me information about the sound level from my microphone. A quick Kagi search revealed several options, but sounddevice seemed like the best one.

The first step is to identify the microphone. For that I need the name as it's know to the operating system. I can get that by running the following code in a Python console:

> import sounddevice as sd
> sd.query_devices()
0 Microsoft Sound Mapper - Input, MME (2 in, 0 out)
1 Microphone (Yeti Stereo Microph, MME (2 in, 0 out)
2 Microphone (WO Mic Device), MME (2 in, 0 out)
....
> sd.query_devices()[1]['name']
'Microphone (Yeti Stereo Microph'

I get a long list of stuff, but I see something with Yeti in the name so I grab that one.

Now let's start listening to the microphone. Sounddevice offers a callback based API, where it passes along the raw audio data received from the microphone. From that, I estimate the loudness by calculating the norm of the sound:

import numpy as np
import sounddevice as sd


def print_sound(indata, frames, t, status):
    volume = np.linalg.norm(indata) * 10
    print(volume)


name = 'Microphone (Yeti Stereo Microph'
with sd.InputStream(device=name,callback=print_sound):
    for i in range(5):
        sd.sleep(1000)

Running this gives something as follows. Can you guess where I snapped my fingers?

0.3724626451730728
0.6015866994857788
0.9348087012767792
0.7427176833152771
0.8615989238023758
0.7162655889987946
0.5638395622372627
0.7117109000682831
59.17434215545654
50.70761203765869
20.951063632965088
14.069621562957764
9.29598331451416
5.908793210983276
3.782018721103668
2.402055263519287
1.7902085185050964
1.1522774398326874
0.793280228972435

The next step is to make it warn me when I speak too loud. For this I keep a buffer of the latest sound intensities in order to be able to detect when either something loud has been happening for a long time or if a really loud noise happened in the last frames:

import time
from collections import deque

import numpy as np
import sounddevice as sd


last_alert = time.time() - 10
q = deque(maxlen=200)


def print_sound(indata, frames, t, status):
    global last_alert
    volume_norm = np.linalg.norm(indata) * 10
    q.append(volume_norm)
    last_elements = [q[i] for i in range(-min(50, len(q)), 0)]
    recent_avg_sound = sum(last_elements) / len(last_elements)
    num_high_count = len([x for x in q if x > 20])
    if num_high_count > 30 or recent_avg_sound > 50:
        if time.time() - last_alert > 10:
            print(f"You are speaking at {volume_norm:.2f}. Think of Gloria!\a")
            last_alert = time.time()


name = 'Microphone (Yeti Stereo Microph'
with sd.InputStream(device=name,callback=print_sound):
    while True:
        sd.sleep(1000)

Now, when running from a Terminal (either on Windows or Linux), this will make a bell sound if in the last 5 seconds (that's about 200 frames) there have been more than 30 frames with loudness over 20 or if in the last second the average was over 50 (this would mean a really loud sound).

If you want to run this outside of Terminal, you can use beepy for example to make sounds and replace the print statement with this:

from beepy import beep

beep(sound="error")

To run this on startup on Windows, I created the following ps1 script in the startup folder:

C:\Users\Roland\Programming\mic_check\.venv\Scripts\pythonw.exe C:\Users\Roland\Programming\mic_check\mic_check.py

Another improvement would be to make it easier to see current loudness and to be able to quit it easily (because the ps1 script runs in the background). For this I used the infi.systray library on Windows:

from infi.systray import SysTrayIcon

def quit(systray):
    global still_on
    still_on = False


menu_options = ()
systray = SysTrayIcon("icon.ico", "Mic check tray icon", menu_options, on_quit=quit)
systray.start()
still_on = True
name = 'Microphone (Yeti Stereo Microph'
with sd.InputStream(device=name,callback=print_sound):
    while still_on:
        sd.sleep(1000)

And now, hopefully I'll learn to control my loudness better!

You can find the full code here.

My next project where I learn in public is about doing transcripts with Whisper.

Whisper is a quite good automatic speech recognition model that is open source and can run on your own computers, provided you have a GPU.

I prefer reading to listening, so I wanted to transcribe my long list of things to listen to, ideally in automatic way.

The first step was to use Whisper to transcribe anything dropped into a folder. I recorded myself while doing this and I plan to do at least 2 more sessions of this.

Lesson learned: OBS studio uses a lot of resources and sometimes the recording has issues because of Whisper also hogging up resources.

The last couple of months have been quite busy for me. Consultant life is starting to pick up the pace (at least some aspects of it), especially the traveling part.

It started in March: Make IT in Oradea, a startup incubator where I'm involved as a mentor, had its first in person pitching event in Timisoara. I was invited to be a part of the jury evaluating the submissions. I had never been to Timisoara before so it was I managed to scratch that itch of traveling to a new city. Unfortunately, I was not very impressed by the city, but at least I met some interesting people.

In April, Today Software Magazine had its first in person event in two years and I presented briefly about modern NLP methods. The presentation was an exercise of my improvisation skills, because the platform used to show the slides had technical difficulties while I was speaking, so I had to wing it.

And then June was busy. I gave a talk at The Developers, the first in person technical conference in Cluj. I talked about Large Language Models. I listened to some great talks and I talked to some people who are also in the consulting business and got some great advice (and some not so encouraging things, such as the guy who signed a contract with a client 7 years after the client heard him speak at a conference).

And then the fun started: flying abroad again. It's been almost two years since I flew and even that was a short 50 minute flight. And it's been 3 years since I have been abroad (Debrecen doesn't really count).

Through Oradea Tech Hub I was invited to mentor at the CodeX hackathon in Riga. It was a really well organized hackathon. The venue was great and it was great to feel the energy from the students (mostly). I definitely wouldn't pull an all-nighter for a hackathon anymore :))

I have to admit I didn't know much about Riga before and I was very pleasantly surprised. It's not a very big city, but it has a cool vibe and it's very clean. 10/10 would recommend, I want to go there again with Roda and Gloria.

It has many buildings in the style of Art Nouveau, like Oradea (where I live), so I finally found out what Art Nouveau is! It also has a statue of St. Roland, so that might have contributed to me liking the city so much.

The last trip in June was to Berlin, where I attended the datalift summit and I was a session chair. That was a really good conference! I met so many great people, all doing machine learning. I hope I can even collaborate with some of them sometime soon.

But the city itself... didn't impress me. I won't go again to Berlin anytime soon. The food was really good, with lots of very diverse options available, but the city itself... meh, too dirty. Sure, the city center is nice, but once you've seen one European capital, you've seen them all.

Now it's time for a bit of relaxing, so now I'm in Sibiu (where I have grandparents). It's Gloria's first longer trip away from home, so we're curious how she'll handle it! So far so good, so we're hoping for at least two more trips in the next months!

My two brothers in law are fans of horse riding and last week they invited the rest of the family to join them for a day of horse riding. They encouraged us with fun tales of how of them fell off/was thrown off a horse. Having not done anything new in a while, Roda and I agreed to it.

When we arrived at the Kalota Lovarda, we received a brief intro to how to ride a horse while it walks and an even shorter intro on how to ride it while it trots. Turns out riding a horse is not a passive activity, but you have to move your hips to nudge the horse to keep moving, while also lightly pulling on the reins. If you let the reins go, the horse will most likely take a break to eat some grass. Horses really love eating grass and will do that whenever they have the chance.

The first five minutes of the intro had me regretting that I signed up for activity. Do this for the rest of the day? Oh boy, oh boy. Please don't fall of the horse! Please don't throw me off, horsey!

It was an interesting experience where I learned a lot, such as the fact that horses have thick skin, so they don't care about going through bushes and thorns - but you do. Also, when going through a thicker forest, horses only make sure they don't hit trees - your knees are none of their concern, or worse, not even your head when going below some low branches.

The views were beautiful, as always, but honestly, I'm not sure if I'd rather do the trips on horses or while walking. There is a certain serenity that comes with the rhythm of horses step, which makes the scenery more enjoyable in a way. But it's so exhausting after 8 hours....

Also, horses have very different personalities and behaviors. Some horses have no problem pooping while walking, others, like my Titi, would always stop whenever he felt the need to relieve himself. Sometimes he would even go off the path, as a courtesy to the others. Some horses would avoid stepping into the mess others left, other horses had no problems marching straight through.

One of the funnier things on the journey was when the owner of the horses got off this horse for some reason and then his horse ran away. I think he spent 15 minutes trying to get back on him. It was quite funny, but it also made most of us reconsider getting off our horses.

When we got to the restaurant after 3 hours, everyone was relieved to finally get off the horse and give our bottoms a well deserved rest. But then it was time to go back, on a slightly different route. This time, even the horses were tired. The older ones would stop periodically for a quick break. But, after 2.5 hours, we did get back to the stables safely.

Unfortunately, I never managed to reliably get my horse to trot and more importantly, I never managed to learn how I should move in the saddle while trotting, to avoid making an omelette. Maybe next time? But it won't be anytime soon. I'd rather go to the aquapark.

While I am a professional software developer (or machine learning consultant, to be more exact) and I have started learning programming more than 10 years ago, I went to college to study it, I spend a lot of my free time learning more about this domain (because I love it), I'm starting to realize that knowing a bit of programming is a super power, from which many people could benefit, if only they knew just a bit of coding.

For example, about 6 years ago I wrote a 2 line JavaScript bookmarklet for a recruiter, that reduced the number of clicks he had to do, from 10 clicks to 3, for each of several thousand candidates. It was a simple script that looked for an HTML tag, extracted a value and put it in the clipboard (from where he pasted it into a spreadsheet). The guy was so happy that he gave me a peer bonus, because it saved him hours of boring work.

Another example is when I have to replace something many times in a text document and I can quickly write a regex to do it. My dad sometimes calls me to do that for him when he is typesetting books.

This week I wanted to read a transcribed sermon that came in a .doc format. For more pleasant reading, I copy pasted it into a markdown file, quickly fixed some Markdown issues (mostly lists and headers), ran it through Pandoc and got a beautiful HTML, that was easy on the eyes. I sprinkled in some copy pasted JavaScript and I had automatic popups for Bible verses when hovering over references.

Julia Evans recently wrote about some tiny personal programs she wrote that both had marginal utility in her life and also brought her joy while writing them.

Doing this kind of programming doesn't require knowing a lot about programming, much less how to develop software professionally, but it does require you to have a knowledge of what is possible with code and either a low enough friction to get started (such as a text editor with a built-in regex search and replace), or a high enough motivation (such as wanting to avoid tens of hours of boring work) to do it.

But unfortunately, I don't know if this is something that can be learned quickly. All these examples I gave took me 5-10 minutes to implement, but that's because I already knew a fair bit of JavaScript, I've had to process Markdown in the past, I've written regexes many times for work and I've had a general awareness of Pandoc, even if I didn't use it much before. If I had to search Kagi for all of these (or if I didn't even know what to search for), it would take a lot longer.

In my opinion, this approach is much more useful than the no-code approach that is popular today (at least among startups) and it would be great to see more startups trying to reduce the friction for doing just a bit of coding, not eliminating coding at all. Grist seems like a pretty good approach in this direction.

What's your favorite quick "hack" where you used a bit of coding knowledge to bring quality of life improvements for yourself?

After 8 years, I finally went again to a Code Retreat, but this time as a facilitator, not as a participant. As far as I know, it was the first time it was done in Oradea, so kudos to the Oradea Tech Hub team for organizing it.

My co-facilitator was Bogdan Bota, who is the co-founder of OptiOffer, with whom I share a surprising number of views on technology and programming. Some of these common things are that neither of us is a big fans of OOP and TDD, so we didn't push that angle too much.

One surprise for me was that the mix of languages that people knew has changed a lot. Java and C# were much rarer, while Javascript was ubiquitous. There were a couple of high school students, who were working with C++ (because that's what's taught in Romania in programming classes).

The participants enjoyed the different challenges we gave them, even if some of them were annoying (but realistic), such as changing requirements or partners mid-session. Bogdi and I had the most fun though, guiding them through this.

We got really good feedback, people had fun and said they learned a lot, and almost everyone was asking about when is the next event going to be. Naomi, I hope you'll organize many more great events!

During the last 1 year I've had a lot of changes in my life. I became an independent machine learning consultant. Other people consider me an entrepreneur. I'm still in disbelief about that label, but it's becoming true, especially since I launched my first side-project (or is still a side project if it's just one of the many things you do?). Sometimes I feel like I barely recognize myself and how I think.

The machine learning consultant part has changed a lot as well. When I started off a year ago, I had many ideas of what I would do: trainings, online courses, consulting, train lots of ML models. Drawing the line at the end of the year has revealed that I did few of the things I thought I would do and that actually, the most profitable things were things I never thought I would do. I barely did any teaching, I did some consulting, but developing a full stack proof of concept for a startup was by far the best thing financially (and really fun too). Hmmmm.... maybe I should update the list of services I offer.

I started out as a machine learning consultant. But almost all of my work has been related to natural language processing (aka: working with text). Hmmmm.... maybe I should niche down to just NLP.

Some things didn't change. I still have a pathological fear of picking up the phone to call someone. Luckily, most of my clients came to me, instead of me having to go to them.

I thought I would have to do a lot of marketing, lots of blog posts, tweets, Youtube videos. I had one good blog post that brought in two clients. All the other clients came through referral and word of mouth. Hmmmm.... who knew networking is that important? And maybe, just maybe, my blog is not as important.

For a long time I was a lurker on forums, rarely posting anything. This year I've discovered several communities where for some reason, I started being more active. Hmmm... maybe I'm not that introverted.

The biggest change was becoming a daddy. I now have a very different reason to look forward to every day: Gloria. Of course, it comes with it's own challenges, including some curveballs, but boy, is it great.

There are some negative changes as well: it was the first year in a long time when I didn't fly at all :(

Here's to more changes in 2020 v2 and to being nimble, as God leads me!