DEV Community: Guy Torbet

How To Learn Machine Learning: For Beginners, by a Beginner

Guy Torbet — Mon, 03 May 2021 18:29:31 +0000

Machine learning and AI as a whole can seem hugely daunting when you're first getting started. Over a weekend, I painstakingly sifted through all of the "beginner" guides on how to get started, so that you don't have to.

I also just made a Twitter! Give it a follow, it'll be worth it I promise ;)

If you like this post, feel free to subscribe or check out my other posts here

Most people recommend an image classifier using the MNIST dataset as your 'hello world'/first machine learning program - I'm not sure if there is just an obscenely high barrier to entry, or if I'm simply just stupid, but for someone who wants to go from zero to something in the world of machine learning, this project seems a bit tricky.

Now of course I attempted it. I read anything and everything that I could get my hands on about the inner workings for how neural networks ACTUALLY work, and thought I had a pretty rudimentary understanding of what was going on, but with pre-made datasets and perplexing technical jargon, it's often hard to dilute what is actually happening - I'm still not entirely sure what the fuck a 2D convolutional layer is.

So without further ado, here's how a beginner can get started with machine learning, by a REAL beginner. This will cover everything, from what libraries and frameworks to use, to how to save and load your trained model. By the end of this project, you should have a rough idea as to how it all works, which will send you on your merry way to more complex projects.

What Framework to Use

There are countless machine learning frameworks and libraries out there, but the main players are TensorFlow, PyTorch and Keras (now part of TensorFlow). As usual, I ended up in a state of paralysis as to which to pick, and which one was the best - this is what inspired me to adopt the mentality of just starting.

In the end, I went with PyTorch. I would recommend this as you get control of everything and it makes it easy to visualise what's going on, but in the grand scheme of things they all work well and it doesn't really matter which one you go with - so just pick one and get started!

First Project

Neural Networks function around arrays, or tensors if you're cool (yes, I know that there is a difference, but it's not that important at this stage).

Long story short, you give a trained model an input as an array, and it gives you an output, that is hopefully correct. The simplest way that I could think to demonstrate this was to count the number of occurrences in a list.

The end goal of this model is to give it a binary list, and it outputs the number of ones, for example;

input: [1, 0, 0, 1, 0, 0, 0, 0, 1, 1], output: 4

Neural networks require an input and an expected output in order to train. You can generate arrays of random 1's and 0's like so, which also includes labels (expected outputs) for easy training later down the line:

def genData(vol, length):

    values = []
    labels = []

        data = []

    for i in range(vol):

        rand = random.randint(0, length)
        data = rand * [1] + (length - rand) * [0]
        random.shuffle(data)

        labels.append(data.count(1))
        values.append(data)

    return values, labels

We then can assign the values and labels to variables, which we convert to tensors so that the neural network can easily read and process them:

x, y = genData(100, 10)

xTensor = torch.Tensor(np.array(x))
yTensor = torch.Tensor(np.array(y))

Simple so far right? Now we structure our network! We will use PyTorches nn.Module to include all of the layers we could ever need, and then just piece them together. For this network, I only used 1 layer, it takes in 10 inputs (I used a random binary array of length 10), and 1 output (0-10, number of 1's in the given list).

We then have a forward method, which tells the network how to process the inputs as they pass through the layers, we're keeping it simple by just passing them through layer 1 and using a relu function:

class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.l1 = nn.Linear(10, 1)

    def forward(self, x):
        out = F.relu(self.l1(x))
        return out

model = Net()

Now, we need a loss function, which will tell our network how far off the expected output it was, so that it can correct itself and become more accurate, we will also add an optimiser, to you know... optimise this process:

criterion = torch.nn.MSELoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

You can do a lot of reading for yourself on the different loss functions, there are hundreds of them, but I enjoyed the MSE or cross entropy loss function for linear NN's.

Now, TRAIN, TRAIN, TRAIN! Think of an epoch as a training loop, so in this example, I'm cycling through the data 1000 times:

for epoch in (t := trange(1000)):
    for i in range(len(x)):
        values = xTensor
        labels = yTensor

        output = model(values)

        optimizer.zero_grad()

        loss = criterion(output, labels.unsqueeze(1))
        loss.backward()

        optimizer.step()

We can finally save the model like this:

torch.save(model, "cnn.pt")

That's it! you've successfully created and trained your first neural network, from the ground up!

Running the Model

Now that we've trained and saved our model, we can start to pass in real inputs, and see what it predicts!

Make a new file, and initialise a simple array, and turn it into a tensor like before:

in_data = [1, 1, 1, 0, 1, 1, 1, 0, 0, 0]
inTensor = torch.Tensor(np.array(in_data))

Next, we need to copy the model from our previous file, so that we can add the weights and biases from our saved neural network to it:

class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.l1 = nn.Linear(10, 1)

    def forward(self, x):
        x = F.relu(self.l1(x))
        return x

Finally, we can load our saved model, evaluate it, and pass in our new input data!

model = torch.load('cnn.pt')
model.eval()

out = model(inTensor)

If you print the value of the "out" variable, you should see a 6 in this case!

If you've made it this far and have been following along, you should hopefully have a better understanding as to how PyTorch, and machine learning, works as a whole. For me, this project was a good jumping off point, and after it I felt more comfortable pursuing image classification with MNIST!

How to Build a Blog in NextJS and Get a Perfect Lighthouse Score

Guy Torbet — Sat, 01 May 2021 16:43:26 +0000

I think that everyone should have a personal site, even if you're not in the development scene - and it doesn't always have to be a massive, expensive headache to set up. While this post is aimed at developers, it can still show you have to make site from the ground up, and completely for free!

I also just made a Twitter! Give it a follow!

If you like this post, feel free to subscribe or check out my other posts here

While at present, it may not have much traffic, it is optimised as fuck - 100% lighthouse scores, quick load times, almost instantaneous indexing by Google, the whole shebang.

Tech stack used:

React with NextJS
Hosting on Vercel
Custom CSS styling
Posts exported from Notion, using gray-matter for metadata
Implementation of the GitHub API

I was torn between create-react-app and NextJS, but as a general rule of thumb, CRA is great for purely single page apps, whereas next is more appropriate for full on websites. A perk of using this is the fantastic hosting on Vercel, which automatically redeploys on a new git commit. This means that as soon as I push a post to the sites repository, Vercel updates the live site, no fuss.

NextJS also offers server side rendering, which at surface level seems ideal for blogs, but the content on my site was simply not dynamically changing enough to warrant the slightly longer load time associated with SSR, so static seems like the best option - for now.

I'm not a designer, far from it - but in my opinion, the site looks pretty clean and minimal. It's done with simple CSS, with some quirky features copied from some CodePens that I found, notably the animated underlines. I didn't have any relevant images, which makes design surprisingly hard, so I decided to focus more on typography. The whole goal of the site was to be concise and to the point, so the design should reflect this.

As for the posts, I write them in Notion, which has a handy feature that allows you to export pages as markdown. ReactMarkdown handles the rendering on the page from there. I then use an NPM package called gray-matter to assign each post with metadata (title, date, description) for easy formatting in React. This also allows me to effectively and automatically update the RSS feed, and is overall the biggest time saver I've found for managing posts!

I'm about to jump into how I handled some of the main features of the site, if you're not interested in the nerd shit then you can skip to the end to see how to optimise your site for Google.

This is how I get all of the posts, it's designed for use in React so the syntax might seem a bit weird:

export async function getPosts() {
        const posts = []
    const context = require.context('../posts', false, /\.md$/)

    for (const key of context.keys()) {
        const post = key.slice(2);
        const content = await import(`../posts/${post}`);
        const meta = matter(content.default)
        posts.push({
            slug: post.replace('.md',''),
            title: meta.data.title,
            date: meta.data.date,
            description: meta.data.description,
        })
    }
    return posts;
}

You can see that it returns a list of objects, each with all the required data to display the posts information and route to it. I can then display them in a simple list like this:

function Posts({ allPosts }) {

    const router = useRouter()

    return (
        <div id={styles.container}>
            <ul>
                {allPosts.map(function (post, key) {
                    return (
                        <li onClick={() => { router.push(`/posts/${post.slug}`) }} key={key} className={styles.tableRow}>
                            <div className={styles.titleContainer}><h2 className={styles.title}>{post.title}</h2></div>
                            <h3 className={styles.date}>{formatDate(post.date)}</h3>
                        </li>
                    )
                })}
            </ul>
        </div>
    )
}

Implementing a RSS feed required the same getPosts function, but in pure nodeJS, here's what that looks like - it's a bit easier to understand what's happening:

function getPosts() {
    var posts = []
    let filenames = fs.readdirSync("posts");

    filenames.forEach((file) => {
        const str = fs.readFileSync(`posts/${file}`, 'utf8');
        const meta = matter(str);
        posts.push({
            slug: file.replace('.md', ''),
            title: meta.data.title,
            date: meta.data.date,
            description: meta.data.description,
        })
    });
    return (posts)
}

I can then fetch specific posts based on the URL, this object also contains the content of the post, which is rendered to the screen, as well as the other meta-data to be passed as props to the post element. The description is only used for the meta tags for improved SEO:

export async function getPostBySlug(slug){
    const fileContent = await import(`../posts/${slug}.md`)
    const meta = matter(fileContent.default)
    const post = meta.data
    const content = meta.content
    return {
        title: post.title,
        date: post.date,
        description: post.description,
        content: content
    }
}

It's pretty simple stuff, which is why I love working with React so much.

NextJS handles dynamic routing amazingly, this is my [slug].jsx. The post data gets passed to a very simple post layout using React's getStaticProps method like so:

function Post(props) {

    return (
        <PostLayout title={props.title} date={formatDate(props.date)} description={props.description}>
            <div id={styles.post}>
                <ReactMarkdown source={props.content} renderers={{code: CodeBlock, image: Image}} />
            </div>
        </PostLayout>
    )
}

export async function getStaticProps(context){
    return {
        props: await getPostBySlug(context.params.slug)
    }
}

export async function getStaticPaths(){
    let paths = await getPosts()
    paths = paths.map(post => ({
        params: { slug:post.slug }
    }));
    return {
        paths: paths,
        fallback: false
    }
}

With the post layout component working like so:

export default function PostLayout(props) {
    return (
        <main>
            <Head>
                <title>{`${props.title} | Guy Torbet`}</title>
                <meta name='description' content={props.description} />
                <meta property="og:title" content={`${props.title} | Guy Torbet`} />
                <meta property="og:description" content={props.description} />
                <meta name="viewport" content="width=device-width, initial-scale=1" />
                <meta charSet="utf-8" />
            </Head>
            <div id={styles.main}>
                <Header title={props.title} />

                <article id={styles.content}>
                    <div id={styles.date}>{props.date}</div>
                    <div id={styles.post}>
                        {props.children}
                        <Mail />
                         </div>
                </article>

            </div>
        </main>
    )

Projects Integration

GitHub have a nice web API to allow for you to access all of a users public repository information, you can see what mine returns here.

This is pretty much everything for my projects section, I still need to tidy up some of the variable names:

function createData(Title, Description, Language, url) {
    return { Title, Description, Language, url };
}

export default function Repos(props) {
    const [arrayItems, setArrayItems] = useState([])
    useEffect(async () => {
        await fetch('https://api.github.com/users/torbet/repos').then(response => response.json())
            .then(data => {
                const items = data.map((arraySingleItem) => {
                    return (
                        createData(arraySingleItem.name, arraySingleItem.description, arraySingleItem.language, arraySingleItem.html_url)
                    );
                });
                setArrayItems(items);
            })

    }, [])

    const router = useRouter()

    return (
        <div id={styles.container}>
            <ul>
                {arrayItems.map((row) => (
                        <li className={styles.tableRow} onClick={() => { router.push(row.url) }} key={row.Title}>
                            <h2 className={styles.title}>{row.Title}</h2>
                            <h3 className={styles.subTitle}>{row.Description}</h3>
                            <h3 className={styles.subTitle}>{row.Language}</h3>
                        </li>
                ))}
            </ul>

        </div>

    );
}

useEffect ensures that the api is only called once every time the page is loaded, this maintains speed whilst ensuring that the page is always up to date.

I love this feature, it means that as soon as I start a new project, it instantly shows up on my site, formatted with the rest of them.

Optimising - Google Lighthouse

It's hard to know when you're done working on project, so I gave myself the target of a perfect lighthouse score. I wouldn't recommend this as you may have to change some fundamental parts of your site to comply with the standards, but in my case, it was simply a case of removing unused javascript, adding a few aria labels and changing some colours for better contrast.

Maybe I'm a web developing prodigy, just a natural talent - more likely, my site was just simple and I had to jump through the loopholes of SEO, but if Google tells me that this is what I have to do, what choice did I have? (plus you get those nice fireworks)

Google Lighthouse is an open-source, automated tool for measuring the quality of web pages.

Lighthouse ranks in 5 main categories:

Performance: This how fast the page is, it measures the global performance. With the important metrics being the First Meaningful & Contentful Paints, the Time To Interactive and the Speed Index.
Accessibility: This measures how accessible your page is. It performs various checks on the page's elements, like the *colour contrast a*nd aria label attributes. The audit has to pass a list defined in here.
Best Practices: This is the reliability of your page, it measure how much the best practices defined by the W3C or Google's standards are used and respected. For example, it will check if your page is served via HTTPS and if errors are present in the console.
SEO: This measures how optimised and standardised your page is for search engines. It checks, for example, if the document contains meta tags and semantic titles.
Progressive Web App (PWA): This measures if your website is able to be installable. It has to pass the audit based on the Baseline PWA Checklist. This isn't normally required for most sites, if you want to start making PWAs, search how to make a manifest.json.

You need a low Time To First Byte. This means that the server needs to respond fast. I use CloudFlare and Vercel for my website and static site generation for my pages, which is essentially the same as using full server-side caching. These provide adequate response times while still allowing me full control over proxies, analytics and version control.

You need to optimise your images. Use SVG and compressed JPG's whenever possible. You can use loading=”lazy” on all image tags to reduce the first contentful paint. The total page size to aim for is under a 1000kb, and images often contribute to a large part of this.

Check your bundle size, the less data you send to clients, the better (within reason). NPM packages often have hidden dependancies, driving up total size of your site. There are many tools to track down which modules are taking up the most space, so go through them and make sure you only have ones that you need.

Improving the accessibility of a webpage is often simpler than one might expect. While there are bigger changes you can make to improve the accessibility of a website, most things require less than a few minutes to address.

Here is a list of some simple improvements that you can make to boost your accessibility rating:

Use an appropriate viewport meta tag.
Ensure all images have an appropriate alt attribute.
Prefer semantic HTML, but fall back to ARIA to help identify elements.
Ensure links have a minimum clickable area of 44x44. Mobile users will love you
Use colours with enough contrast to help the readability of text. Most browser dev tools can help you to find the right combination.

Optimising for lighthouse might take a while, but it helps to be methodical. Work in iterations where you:

Target an area.
Optimise.
Measure the results.
GOTO 1.

Websites need constant maintenance to stay relevant and up-to-date. As techniques evolve, and the web progresses, we need to adapt what we have to best serve our visitors. Hopefully this article serves as a checklist for your website, and it can help you achieve a better score which, ultimately, not only gives you a chance at a better placement in search engines, but it also serves your visitors with a better experience.

The Most Valuable Trait in 2021

Guy Torbet — Sat, 01 May 2021 14:37:47 +0000

The word "amateur" has many connotations, beginner, inexperienced, but to me, it’s someone who is doing something for the love of it, and not seeking fame or recognition in return. That’s a huge advantage amateurs have over professionals. They aren’t afraid of experimenting, making mistakes, trying something new. Pick something you WANT to learn, and allow it to consume you.

I just made a Twitter! Give it a follow to stay up to date, it'll be worth it I promise ;)

Also, if you like this post, feel free to subscribe or check out my other writing here

When someone feels genuinely passionate about what they are doing, the overall final outcome, be that an article or product, that they produce is often far superior to those of people who focus solely on said outcome. In essence, focus on the journey, not the destination.

Think back to school, for me, it's not that far back. Think about a subject you disliked, hated, or at least didn't enjoy, and how much effort and turmoil it took to study and learn that subject, purely for the exam grade that came at the end of the year. Now think about your hobby, your passion; you can easily sink hundreds of hours researching and learning about it, why? because it peaks your curiosity.

In my experience, being curious is the single most defining trait that I see in myself and others. It allows you to be comfortable with the unknown and inspires you to understand that which you don't already.

Why Curiosity Is The BEST

The past 100 years did not reward curiosity. Traditional schools, corporations, and churches didn’t just deter open inquiry and experimentation—they feared it. Instead, they rewarded emulation. Any sort of innovation or questioning was limited to a few people at the very tippy top of the food chain, with everyone else expected to be a good little worker bee.

But the internet has inverted this. Today, it seems that it's the ones who fail to experiment, innovate, or challenge who get left behind....

It doesn’t matter what you are trying to become better at, if you only do the work when you’re "motivated", then you’ll never be consistent enough to become a professional. I've realised, that motivation is a product of action, if you get stared on a project/ task/ job, you will quickly become motivated to keep going, and to do better. The ability to show up everyday, stick to the schedule, and do the work — especially when you don’t feel like it — is so valuable that it is literally all you need to become better 99% of the time.

When I don’t miss workouts, I get in the best shape of my life. When I code every day, I become a better programmer. When I write every week, I become a better writer. Having the fundamental enjoyment that comes with being an amateur, and the determination and drive that comes with being curious about what you're doing, catalyses this effect 10 fold.

Despite some people in powers opinion, most of the time, you can and should monetize your hobbies. If you enjoy it, you're gonna be a good at it, and whatever you're good at will have a value associated with it! Put yourself out there, share what you are working on, and who knows: one day it might set you free!

Everything you Need to Know about SafeMoon, the new cool cryptocurrency

Guy Torbet — Sat, 01 May 2021 12:52:58 +0000

I'll preface this by saying that this is not financial advice by any stretch of the imagination, furthermore, I have no idea how this whole crypto thing works, so take everything written here with a grain of salt.

I also just made a Twitter! Give it a follow to stay up to date!

If you like this post, feel free to subscribe or check out my other posts here

I have always wanted to give day trading a shot, but with so many options, it can start to get overwhelming. I attempted it for the first time a few months ago, and was disappointed - most large markets tend to be slow moving, boring, and require a large investment to make any meaningful returns; this is where crypto steps in. Crypto is generally a lot more volatile than traditional stocks and shares, the price of a coin can plummet or soar in a matter of hours - high risk high reward, I like it.

Now, I'm ashamed to admit this, but in an attempt to decrease the endless boredom of lockdown, I caved and downloaded TikTok. Before long, I had made it onto finance recommendations (for whatever reason). I was scrolling meaninglessly until I discovered a video on the new cryptocurrency "SafeMoon".

Initially, I was sceptical - I mean, how many times have you heard "become an early investor in the next Bitcoin", so I did some more digging. SafeMoon is unique in that it rewards holding the coin for longer periods of time, #HODL if you're THAT cool. Every time someone sells their SafeMoon, 5% of it gets reimbursed to current holders, 5% gets burned, and the rest released back into the pool. This means that when people sell their SafeMoon, your amount increases and the value goes up.

The nature of crypto is that there is a finite amount, meaning the laws of supply and demand apply to it; the more that is bought, the lower the supply is, the higher the price. Given that this specific coin was garnering a lot of attention and therefore demand, the price curve already showed tremendous growth, so I decided to give it a try.

My Attempt

I put in £200 initially, and lost a lot of it. This was partly because it was my first time trying out crypto, I had no idea what I was doing, and spent a lot of withdrawal and transfer fees. But also because Guy Torbet apparently has the power to cripple an entire market under the sheer weight of his presence, because as soon as I figured out how to actually GET SafeMoon, the entire crypto market took a massive dip. Luckily it bounced back the next day.

This wasn't the end of the world, I had seen the potential growth of Aragon ($ANT), invested in it, and it took off - mostly covering my losses. This was a pure fluke, and nothing on what I was expecting to make on my new love, SafeMoon.

Now, when crypto dips, what do we do?

A) call it quits, sell it all, give others a free 5% of our holdings and end up losing money.

B) invest even more while the price is low and hope for the best.

If you picked A, shame on you, you have no place in this HODL cult. If you're a B, firstly, you have balls the size of coconuts, and secondly, please don't put in money that you cant afford to lose.

Now, a major reason why SafeMoon appealed was because of how new it is, it was developed just a few weeks ago, and hence, is not on the open-market or exchanges (yet). This will change in the coming weeks when it gets released more publicly, and should cause the price to skyrocket due to the increased demand for the coin.

As I became more confident in how crypto works, I felt more comfortable sinking a few more dollars into it. There was a huge dip in Polkadot, so I bought some for cheap, and the price has been rising steadily ever since.

There is definitely some safer options (ironically), and crypto can be a very worthy long time investment, but I'm more a low effort big rewards right now kinda guy, so, I said fuck it, and went for it.

How to Buy SafeMoon

Buy Binance coin ($BNB), I originally used Crypto.com to do this, but bailed due to their extortionate withdrawal fees, I recommend Binance themselves for this, fuck crypto.com.
Make a Trust Wallet account and transfer your BNB into it.
Swap your coins to BNB smart chain in the Trust Wallet app.
Open up pancake swap, connect your Trust wallet, and swap your Binance coins for SafeMoon. It is not officially on the PancakeSwap list, so enter the token (0x8076C74C5e3F5852037F31Ff0093Eeb8c8ADd8D3) or go directly here.

Additional steps: apparently Binance in the US is super duper shit, so you may need to find an alternative exchange to purchase your BNB. Occasionally PancakeSwap will drop an error when swapping to SafeMoon, click the settings icon and set your slipping rate to 11% (or slightly higher if it doesn't work).

I'm not upset about the losses, the way I see it, I invested in learning something new. SafeMoon, and most cryptocurrencies for that matter, is a massive gamble - but I'm gonna continue to hodl until we react the moon!

I like crypto, and will spend more time understanding it and researching, as it becomes more and more mainstream, the returns should only get bigger!

If you have some money to play around with, I'd put a bit into SafeMoon, but nothing that you can't afford to lose. It has massive potential in all aspects, but if you want a more long term investment, keep being an "A" and put your money in the S&P 500 like a normie.

How to REALLY handle mobile viewports

Guy Torbet — Fri, 16 Apr 2021 15:41:19 +0000

I don't know if you've noticed, but using the vertical viewport on mobile is shit. Here's a quick guide on how to make them less awful.

If you like this post, feel free to subscribe or check out my other posts here

Look at the following picture and tell me where you think 100vh should be. Most of you probably said the green arrow, you know, the VISIBLE bit? Actually, 100vh goes all the way to the red arrow, why? I don't know, maybe because that little menu bar at the bottom is ever so slightly transparent that it counts as 'visible'?

Say you want a 100vh div, just to fill the space of the green arrow and nothing more, if you're on IOS, it's easy, just use webkit to fill the space:

.div {
  min-height: 100vh;
  /* fill on ios
  min-height: -webkit-fill-available;
}

But what if you're on Android? or if you need a more specific viewport size? No fear, I can help with that aswell!

First, get the value of the windows height, then take 1% of that. You can then set a CSS variable (I used —vh) using this value:

let vh = window.innerHeight * 0.01;
 document.documentElement.style.setProperty('--vh', `${vh}px`);

Then, it's as simple as multiplying your variable by how ever many units you want, for example, 50vh equivalent would be:

.div {
  height: calc(var(--vh, 1vh) * 50); // what vh you want
}

But Wait, There's More

If I were you, I'd add this nifty little resize event to change the value of your —vh variable, it will keep your sizing consistent, even when the window size changes.

Just wrap a resize event listener around where you set your variable:

window.addEventListener('resize', () => {
    let vh = window.innerHeight * 0.01;
    document.documentElement.style.setProperty('--vh', `${vh}px`);
  });

I use this whenever I need to set an elements vh, it's consitent across desktop and mobile, and a real time saver!

How I Learned Machine Learning in a Weekend

Guy Torbet — Fri, 09 Apr 2021 19:29:53 +0000

If you like this post, feel free to subscribe or check out my other posts here

What Framework to Use

First Project

Neural Networks function around arrays, or tensors if you're cool (yes, I know that there is a difference, but it's not that important at this stage).

The end goal of this model is to give it a binary list, and it outputs the number of ones, for example;

input: [1, 0, 0, 1, 0, 0, 0, 0, 1, 1], output: 4

def genData(vol, length):

    values = []
    labels = []

        data = []

    for i in range(vol):

        rand = random.randint(0, length)
        data = rand * [1] + (length - rand) * [0]
        random.shuffle(data)

        labels.append(data.count(1))
        values.append(data)

    return values, labels

We then can assign the values and labels to variables, which we convert to tensors so that the neural network can easily read and process them:

x, y = genData(100, 10)

xTensor = torch.Tensor(np.array(x))
yTensor = torch.Tensor(np.array(y))

class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.l1 = nn.Linear(10, 1)

    def forward(self, x):
        out = F.relu(self.l1(x))
        return out

model = Net()

criterion = torch.nn.MSELoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

You can do a lot of reading for yourself on the different loss functions, there are hundreds of them, but I enjoyed the MSE or cross entropy loss function for linear NN's.

Now, TRAIN, TRAIN, TRAIN! Think of an epoch as a training loop, so in this example, I'm cycling through the data 1000 times:

for epoch in (t := trange(1000)):
    for i in range(len(x)):
        values = xTensor
        labels = yTensor

        output = model(values)

        optimizer.zero_grad()

        loss = criterion(output, labels.unsqueeze(1))
        loss.backward()

        optimizer.step()

We can finally save the model like this:

torch.save(model, "cnn.pt")

That's it! you've successfully created and trained your first neural network, from the ground up!

Running the Model

Now that we've trained and saved our model, we can start to pass in real inputs, and see what it predicts!

Make a new file, and initialise a simple array, and turn it into a tensor like before:

in_data = [1, 1, 1, 0, 1, 1, 1, 0, 0, 0]
inTensor = torch.Tensor(np.array(in_data))

Next, we need to copy the model from our previous file, so that we can add the weights and biases from our saved neural network to it:

class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.l1 = nn.Linear(10, 1)

    def forward(self, x):
        x = F.relu(self.l1(x))
        return x

Finally, we can load our saved model, evaluate it, and pass in our new input data!

model = torch.load('cnn.pt')
model.eval()

out = model(inTensor)

If you print the value of the "out" variable, you should see a 6 in this case!

I Asked 100 People What They Learned From Lockdown

Guy Torbet — Sat, 03 Apr 2021 13:29:39 +0000

I like hearing about people's life lessons - the more of them you learn, the more traps and pitfalls you can avoid as you live your life. Covid-19 and subsequently lockdown has been a journey for everyone to say the least, it started off feeling like a holiday, but now, more than a year in, its initial charm has definitely worn off. So, I asked people the most valuable lessons that they have learned from being locked away in their homes, so hopefully you can pick out some juicy nuggets to help you along the way!

If you like this post, feel free to subscribe or check out my other posts here

I've taken the liberty to organise the best responses into sections, for easy reading; feel free to reach out to me if you have learned something that you want featured on the list!

Habits

Consistency is better than anything when it comes to making positive changes. No matter what it is, if done consistently, you can improve.

Having a routine is severely underrated. I never realised just how much I needed one until my overall physical and mental health took a complete nosedive.

The power and importance of habits is heavily emphasised in "Atomic Habits" by James Clear, the ability to make an action 'second nature' and be able to do it subconsciously is something that I have found incredibly useful! (Good book too 😉 )

Overcoming Boredom

That hobbies are key to keeping your sanity, but then again, I never went into lockdown, I worked through it all.

spend time on your hobbies because they will make you happy

Sometimes, being able entertain oneself is far more valuable than that of entertainment provided by others.

I have to say that I was never really "bored" during lockdown, like many others, I took it as an opportunity to get more work done and learn new skills to improve myself. Have a look at the motivation section if you want to kickstart your journey during lockdown.

Productivity

I don’t have enough time is no longer a valid excuse. I had a crazy amount of free time during peak COVID and didn’t utilise it how I should have. Taught me that you can be as productive as you want to be, after all everyone has 24 hours in a day.

Adaptability. Being able to change course more easily and feel less stress about changes to plans. I used to worry a lot more about the unknowns in my tightly controlled life, but having everything flip do quickly and being able to adapt and the world not ending. It definitely has helped me worry less about things outside my control.

Write. Things. Down. It helps keep yourself accountable.

Lockdown has really tested to work ethic and discipline of us all. While some may have crumbled under the vast amounts of 'free time', others made use of the extra hours and put them towards something meaningful. For me, lockdown has been a great test of my willpower when it comes to getting work done, and it's taught me that I can manage myself effectively, even without the structure that school and university provides.

Motivation

Yep, I learned that. And as an extension of that thought - the days where you aren’t motivated and don’t feel like getting to it are the days that really count. Anyone can do something sporadically when they are motivated.

That just because I have more "free time" doesn't mean that I'm going to magically have the motivation to do all the things I've been procrastinating to do beforehand. I gotta be proactive and develop the self discipline to do so every day.

To just do it! I put so many things off because I thought I had time or couldn’t make a decision. What was I waiting for? No need to analyze everything to death. Just do it and figure it out. The other thing I started to practice was talking to strangers more. If someone initiated a conversation I would slow down and have it instead of rushing it. I realized that for some that in person conversation might be the only one they have had in a while. I especially do this at the grocery store. I am also more adaptable than I thought.

To stop procrastinating. I have a really big problem with overthinking and analysis paralysis. I realised this is really affecting me in a negative way.

In order to be happy, you need to sort things out for yourself.

A lot of people seemed to have issues in this department - overthinking, planning paralysis, procrastination, sound familiar? I faced the exact same issues at the beginning, until I realised that the only thing you can do to overcome it is to Just Start.

Work

That working from home is not necessarily a good thing (and is possibly more bad than good).
I was always skeptical about the concept because my main concern was that it would be a breeding ground for fraud. This turned out to be true, but that's another story.
It also happens to be massively detrimental to your productivity. As it turns out, being at home in a place with a lot of distracting things you would rather do instead of work while struggling with procrastination isn't really a recipe for success (I learned this the hard way). That's doubly true when you add screaming children into the mix that constantly want your attention while you are trying to work.
And the cherry on top is now, as far as your employers are concerned, you are NEVER not in your office. If you thought your company calling you late at night because something went wrong at the office was invasive and disrespectful of your time, then wait until you literally can't leave the office. Now they will pull all sorts of shit that they previously weren't able to do.
Of course, you could tell them that it isn't your problem and you could try to set your boundaries, but they will make it your problem and they will push those boundaries every chance they get. Why? because to them, your bedroom belongs to them now.

You can do 90% of the work without going to office if you're working on computers.

All of the work I've ever done has been remote, and while I enjoy it for the most part, I long to experience working side by side with colleagues in an office and learning from each-other. Hopefully, after I finish my degree, I can experience that, without a mask in between us.

Exercise

Stay active even though there was barely a reason to get out. I’m now in serious pain (neck/back) and I’m trying to undo what I’ve done to myself.

there can be no mental health without physical health. It took 2 months of not getting out for my body to start getting random aches and pains and pinched nerves which I absolutely did not expect at my age. And all of that decreased my quality of life tremendously. Now that the weather is nice again, I'm not missing any chances to go outside and be active.

I won't lie, I'm not the most active person. Don't get me wrong, the dog has been VERY well walked for the past year, but asking about peoples relationships with health and exercise has made me realise that i need to place more of an emphasis on this, and my later-self will thank me.

Social

Your support network isn't what you think it is. Cowards will out, and your true people will step in to help in whatever way works for them. Sometimes thats just ordering you food bc you haven't had a full meal in 3days, sometimes its hooking you up with a job referral,
Real family (chosen or otherwise) does what they can for each other, even if it seems small, bc anything is better than nothing.

Don't waste time and energy on people who don't bring anything productive or positive to your life.

I learned just how much surrounding myself with my peers can really boost my mood. It’s important to surround yourself with people just as much as it is to have alone time. I miss seeing people everyday. My friendly exchange in conversation with the 7/11 cashier is the highlight of my day sometimes

That I need social stress to be productive

Noticing how much better I felt not being obligated to see anyone made me really reevaluate my relationships and notice my patterns of choosing toxic people, which I'm now working on in therapy.

Your relationships are what you make of them and can change for the better with lover and persistence

I don't have as many friends as I thought I did.

Sometimes, being able entertain oneself is far more valuable than that of entertainment provided by others. (x2)

All those years of being antisocial and a homebody were just training.

I've been quite content with being by myself, which has played into my hands nicely during quarantine. I've always valued having a small, but stable group of friends to surround myself with, which has been an absolute god-send during the past year.

Self

Be kind to myself. As soon as lockdown started, I fell into the boat that if I don’t come out of lockdown with a new skill or business that I’ve failed. So I learned to love myself more, which in turn has given me a great balance of being productive and just enjoying each day.

Life is hard. No need to make it harder by constantly judging myself. Especially for things that don't hurt anyone and won't matter in the long run.

I have a tendency to be too tough on myself, but it goes further than 'being my own worst critic'. I get genuinely annoyed at myself if I don't get enough work done or put things off until the last minute. Even although that is my fault, I've learned that my standards were simply too high to be maintained consistently, and every now and then I should cut myself some slack.

Gratitude

Not taking anything for granted. We generally tend to realise the true value of simple things once we lose them.

To be grateful for everything I have in life. Simple as that.

Slowing down and enjoying your severance pay is good but having goals for each day makes your rest feel earned.
Be a good neighbor because talking between porches and running into familiar people every day makes the pandemic feel less awful.

Appreciate that you’re able to wake up the next day and be given the chance to breathe in fresh air and have the sun kiss our skin.
Thousands of lives who are no longer with us won’t ever get to wake up again.

Fundamental gratitude is something that I'm actively developing in myself. In the grand scheme of things, we get it easy, and looking at the positives as blessings and realising that the negatives aren't THAT bad, could improve your happiness dramatically

Miscellaneous

Nihilism (in a good way). Society is just a bunch of social constructs stacked on top of each-other, and we're fully capable of changing those constructs at any time. All we have have to do is decide to do it, and it's done. The only barrier is getting people to agree to do it, and they will fight you at all costs, for basically no reason. People couldn't even agree to wear masks during a deadly, highly contagious viral outbreak because it's not what they're used to doing, so imagine the uphill battle were in for when tackling the 40 hour work week.
My mindset has shifted from "how do I prove XYZ is true?" to "how do I convince people to treat XYZ as true?". Because all.the proof in the world doesn't necessarily change behaviour, and what people belief doesn't matter nearly as much as how they behave because holy shit there is no end to the cognitive dissonance that human beings are capable of. The realisation that narcissism can be so severe that it's essentially become its own variation of psychosis is astounding, but helpful for realising where to stop wasting my time and where we need to actually focus our efforts.

I've learned emacs.

Red wine can replace orange juice with breakfast and it has a longer shelf life.....okay I learned this 20 years ago but it's still true today.

toilet paper is not scarce

So, What have I learned? First lesson: Never attempt to eat a bat, secondly - the power of acceptance, don't stress about what's out of your control, sometimes it just "is what it is".

I used to fixate over little things that now seem so pointless and small when you consider the bigger picture. I re-read 'The Subtle Art of Not Giving a Fuck" by Mark Manson, and second time around, the key ideas really stuck. So now, every time I encounter a problem, I will always try to ask myself now: Is this worth the stress? Will this even matter 1, 5, 10 years from now? I'm 17, still young, the chances that something I do now have any meaningful impact on my future or career are slim to none, so I've started to accept that (sometimes) I fuck up, and just move on.

How I Got Into Crypto With SafeMoon

Guy Torbet — Fri, 26 Mar 2021 15:19:32 +0000

I also just made a Twitter! Give it a follow!

If you like this post, feel free to subscribe or check out my other posts here

My Attempt

Now, when crypto dips, what do we do?

A) call it quits, sell it all, give others a free 5% of our holdings and end up losing money.

B) invest even more while the price is low and hope for the best.

If you picked A, shame on you, you have no place in this HODL cult. If you're a B, firstly, you have balls the size of coconuts, and secondly, please don't put in money that you cant afford to lose.

How to Buy SafeMoon

Buy Binance coin ($BNB), I originally used Crypto.com to do this, but bailed due to their extortionate withdrawal fees, I recommend Binance themselves for this, fuck crypto.com.
Make a Trust Wallet account and transfer your BNB into it.
Swap your coins to BNB smart chain in the Trust Wallet app.
Open up pancake swap, connect your Trust wallet, and swap your Binance coins for SafeMoon. It is not officially on the PancakeSwap list, so enter the token (0x8076C74C5e3F5852037F31Ff0093Eeb8c8ADd8D3) or go directly here.

I like crypto, and will spend more time understanding it and researching, as it becomes more and more mainstream, the returns should only get bigger!

Just Start

Guy Torbet — Sun, 14 Mar 2021 12:42:47 +0000

The best time to plant a tree was twenty years ago. The second best time is now.

No, the second best time was 19 years and 364 days ago - but I get what you mean.

You read all these stories of the early investors in bitcoin, racking in their millions today, the tech billionaires who got the jump on the competition. Its easy to make yourself believe that you have missed the opportunity for greatness, but in todays climate, that could not be further from the truth.

Whatever you want to do in life, the best time to start is now. Waiting for the perfect opportunity or for confirmation that your idea is good will lead to a paralysing state of inaction. You don’t need validation for your idea to get started. Just get to work, see how you feel about it, and keep going if it feels right.

Take this newsletter for example, my instinct is to feel that I should give readers scientific, solid, evidence-based content. Why would anyone care about me and my opinion? The honest answer is - they probably won't, but that's not stopping me. I'm not writing to garner an audience or for profit, I'm writing because I wanted a way to document my journey, so I just started writing.

The Flywheel

This is one of my favourite concepts that I've read about recently after reading "Good to Great" by Jim Collins.

Picture a huge, heavy flywheel—a massive metal disk mounted horizontally on an axle, about 30 feet in diameter, 2 feet thick, and weighing about 5,000 pounds. Now imagine that your task is to get the flywheel rotating on the axle as fast and long as possible. Pushing with great effort, you get the flywheel to inch forward, moving almost imperceptibly at first. You keep pushing and, after two or three hours of persistent effort, you get the flywheel to complete one entire turn. You keep pushing, and the flywheel begins to move a bit faster, and with continued great effort, you move it around a second rotation. You keep pushing in a consistent direction. Three turns ... four ... five ... six ... the flywheel builds up speed ... seven ... eight ... you keep pushing ... nine ... ten ... it builds momentum ... eleven ... twelve ... moving faster with each turn ... twenty ... thirty ... fifty ... a hundred.
Then, at some point—breakthrough! The momentum of the thing kicks in in your favor, hurling the flywheel forward, turn after turn ... whoosh! ... its own heavy weight working for you. You’re pushing no harder than during the first rotation, but the flywheel goes faster and faster. Each turn of the flywheel builds upon work done earlier, compounding your investment of effort. A thousand times faster, then ten thousand, then a hundred thousand. The huge heavy disk flies forward, with almost unstoppable momentum.

It's human nature to look at a successful person or business, and think of a thousand ways in which they could grow their audience / revenue, many of which are easy to implement for them. However - when it comes to going from nothing to something, the idea of growth seems far more aloof.

If you’ve studied chemistry, you’ll know about activation energy - the ‘hump’ of energy required to get a reaction started. It’s the same in real life - if we want to get anything done, we need to first put in a burst of activation energy to begin with, and then things become a lot easier.

To overcome activation energy, you can force yourself to start. Everything takes care of itself after that. I've started telling myself that I'll "only work for 10 minutes", but by the time thats up, I'm too engrossed in what I'm doing to be able to stop.

Building habits and routine is fundamental to productivity and success - in essence; don't think, just do.

All this isn't to say that planning isn't paramount. Don't make a mistake that hundreds have made before you, but there comes a point where no amount of planning will be enough, you'll paralyse yourself.

I recently learned about AI and deep learning, and with so many options for frameworks, I was overwhelmed and frustrated that no one could give me the answer to which was the "best". Eventually, I just said fuck it and picked PyTorch and ran with it, so when it came time to try out another, I had the fundamentals and it was easy to learn!

No amount of planning is a substitute for action, get an idea of what you want to do, write down how you will get there, and figure the rest out on the way!

At the start, stop worrying about the outcome, and just enjoy the journey. After all, "the journey of a thousand miles begins with a single step" - So stop fucking around, and just start.

How I Built My Site (with Perfect Lighthouse Score)

Guy Torbet — Fri, 12 Mar 2021 15:35:02 +0000

I also just made a Twitter! Give it a follow!

If you like this post, feel free to subscribe or check out my other posts here

While at present, it may not have much traffic, it is optimised as fuck - 100% lighthouse scores, quick load times, almost instantaneous indexing by Google, the whole shebang.

Tech stack used:

React with NextJS
Hosting on Vercel
Custom CSS styling
Posts exported from Notion, using gray-matter for metadata
Implementation of the GitHub API

I'm about to jump into how I handled some of the main features of the site, if you're not interested in the nerd shit then you can skip to the end to see how to optimise your site for Google.

This is how I get all of the posts, it's designed for use in React so the syntax might seem a bit weird:

export async function getPosts() {
        const posts = []
    const context = require.context('../posts', false, /\.md$/)

    for (const key of context.keys()) {
        const post = key.slice(2);
        const content = await import(`../posts/${post}`);
        const meta = matter(content.default)
        posts.push({
            slug: post.replace('.md',''),
            title: meta.data.title,
            date: meta.data.date,
            description: meta.data.description,
        })
    }
    return posts;
}

You can see that it returns a list of objects, each with all the required data to display the posts information and route to it. I can then display them in a simple list like this:

function Posts({ allPosts }) {

    const router = useRouter()

    return (
        <div id={styles.container}>
            <ul>
                {allPosts.map(function (post, key) {
                    return (
                        <li onClick={() => { router.push(`/posts/${post.slug}`) }} key={key} className={styles.tableRow}>
                            <div className={styles.titleContainer}><h2 className={styles.title}>{post.title}</h2></div>
                            <h3 className={styles.date}>{formatDate(post.date)}</h3>
                        </li>
                    )
                })}
            </ul>
        </div>
    )
}

Implementing a RSS feed required the same getPosts function, but in pure nodeJS, here's what that looks like - it's a bit easier to understand what's happening:

function getPosts() {
    var posts = []
    let filenames = fs.readdirSync("posts");

    filenames.forEach((file) => {
        const str = fs.readFileSync(`posts/${file}`, 'utf8');
        const meta = matter(str);
        posts.push({
            slug: file.replace('.md', ''),
            title: meta.data.title,
            date: meta.data.date,
            description: meta.data.description,
        })
    });
    return (posts)
}

export async function getPostBySlug(slug){
    const fileContent = await import(`../posts/${slug}.md`)
    const meta = matter(fileContent.default)
    const post = meta.data
    const content = meta.content
    return {
        title: post.title,
        date: post.date,
        description: post.description,
        content: content
    }
}

It's pretty simple stuff, which is why I love working with React so much.

NextJS handles dynamic routing amazingly, this is my [slug].jsx. The post data gets passed to a very simple post layout using React's getStaticProps method like so:

function Post(props) {

    return (
        <PostLayout title={props.title} date={formatDate(props.date)} description={props.description}>
            <div id={styles.post}>
                <ReactMarkdown source={props.content} renderers={{code: CodeBlock, image: Image}} />
            </div>
        </PostLayout>
    )
}

export async function getStaticProps(context){
    return {
        props: await getPostBySlug(context.params.slug)
    }
}

export async function getStaticPaths(){
    let paths = await getPosts()
    paths = paths.map(post => ({
        params: { slug:post.slug }
    }));
    return {
        paths: paths,
        fallback: false
    }
}

With the post layout component working like so:

export default function PostLayout(props) {
    return (
        <main>
            <Head>
                <title>{`${props.title} | Guy Torbet`}</title>
                <meta name='description' content={props.description} />
                <meta property="og:title" content={`${props.title} | Guy Torbet`} />
                <meta property="og:description" content={props.description} />
                <meta name="viewport" content="width=device-width, initial-scale=1" />
                <meta charSet="utf-8" />
            </Head>
            <div id={styles.main}>
                <Header title={props.title} />

                <article id={styles.content}>
                    <div id={styles.date}>{props.date}</div>
                    <div id={styles.post}>
                        {props.children}
                        <Mail />
                         </div>
                </article>

            </div>
        </main>
    )

Projects Integration

GitHub have a nice web API to allow for you to access all of a users public repository information, you can see what mine returns here.

This is pretty much everything for my projects section, I still need to tidy up some of the variable names:

function createData(Title, Description, Language, url) {
    return { Title, Description, Language, url };
}

export default function Repos(props) {
    const [arrayItems, setArrayItems] = useState([])
    useEffect(async () => {
        await fetch('https://api.github.com/users/torbet/repos').then(response => response.json())
            .then(data => {
                const items = data.map((arraySingleItem) => {
                    return (
                        createData(arraySingleItem.name, arraySingleItem.description, arraySingleItem.language, arraySingleItem.html_url)
                    );
                });
                setArrayItems(items);
            })

    }, [])

    const router = useRouter()

    return (
        <div id={styles.container}>
            <ul>
                {arrayItems.map((row) => (
                        <li className={styles.tableRow} onClick={() => { router.push(row.url) }} key={row.Title}>
                            <h2 className={styles.title}>{row.Title}</h2>
                            <h3 className={styles.subTitle}>{row.Description}</h3>
                            <h3 className={styles.subTitle}>{row.Language}</h3>
                        </li>
                ))}
            </ul>

        </div>

    );
}

useEffect ensures that the api is only called once every time the page is loaded, this maintains speed whilst ensuring that the page is always up to date.

I love this feature, it means that as soon as I start a new project, it instantly shows up on my site, formatted with the rest of them.

Optimising - Google Lighthouse

Google Lighthouse is an open-source, automated tool for measuring the quality of web pages.

Lighthouse ranks in 5 main categories:

Performance: This how fast the page is, it measures the global performance. With the important metrics being the First Meaningful & Contentful Paints, the Time To Interactive and the Speed Index.
Accessibility: This measures how accessible your page is. It performs various checks on the page's elements, like the *colour contrast a*nd aria label attributes. The audit has to pass a list defined in here.
Best Practices: This is the reliability of your page, it measure how much the best practices defined by the W3C or Google's standards are used and respected. For example, it will check if your page is served via HTTPS and if errors are present in the console.
SEO: This measures how optimised and standardised your page is for search engines. It checks, for example, if the document contains meta tags and semantic titles.
Progressive Web App (PWA): This measures if your website is able to be installable. It has to pass the audit based on the Baseline PWA Checklist. This isn't normally required for most sites, if you want to start making PWAs, search how to make a manifest.json.

Here is a list of some simple improvements that you can make to boost your accessibility rating:

Use an appropriate viewport meta tag.
Ensure all images have an appropriate alt attribute.
Prefer semantic HTML, but fall back to ARIA to help identify elements.
Ensure links have a minimum clickable area of 44x44. Mobile users will love you
Use colours with enough contrast to help the readability of text. Most browser dev tools can help you to find the right combination.

Optimising for lighthouse might take a while, but it helps to be methodical. Work in iterations where you:

Target an area.
Optimise.
Measure the results.
GOTO 1.