Hummingbot Community, 29 Jul 2023

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: Hummingbot Community Call - v1.18, Bounties Updates, Directional Strategy Experiment Demo

Description

Community Call Agenda:
• v1.18 Release Highlights
• Bounties Update
• Directional Strategy Experiment Demo

00:00 v1.18 Release Highlights
6:04 Community Bounties Update
10:39 Directional Strategy Experiment Lecture
23:31 Results
24:47 Questions / Discussions

A

All right, uh so, let's go ahead and get started with the community call for this month. But first uh let me go over the do changes we introduced in the last release. Five different repositories within the open source code base, uh one of which is hummingbot. uh Another is Gateway our desk connector module, uh and then we also have a deploy example, repo that ties everything.

B

Together and provides, you know, standardized ways.

A

Where people install different configurations uh using Docker, in addition, we're also working on a dashboard repo which is going to be more of a community project that allows people to have more of a companion app that they use to create and manage amount instances. If you're kind of new hummingbot easiest way to get started is accessing the.

B

Deploy examples.

A

And then picking the configuration you want and then just running Docker compose up what I know is that we are fixing an ongoing issue with the Gateway module. We recently moved the token lists out into to the conf directory to allow users to go to edit and add tokens. However, that seems to have some issues, especially when users are trying to update Gateway and compliant. Sometimes token lookup may not be working, and if you encounter the issue, it's restarting both Gateway and client, we should fix it.

A

uh This is an ongoing issue, uh we're still trying to track down the root cause, but we're pretty confident we can get this fixed pretty soon.

A

So one of the biggest changes in version 1.18 was the edition of python 310.. Before this hummingbot was built using python 3.8, our CTO Fede spent a lot of time in the last few weeks. You know making sure that all the features in hummingbot and all the connectors were worked with this upgrade. In addition, uh this upgrade should improving performance and reliability.

A

It also allows users who are using that new Apple M2 uh chipset to use honeybot, whereas before it wasn't supported another change that was pretty big, was in the Gateway repo Victor from coin Alpha. uh He took bounties and refactored every single chain endpoint um into a set of standardized ones, and we think we think that going forward, every new connector should override these existing endpoints or propose new ones. However, there should be no, like you know, like uh you know, ethereum slash approve, or you know slash.

A

You know, kujira slash tokens that kind of brow structure. It's not gonna do.

B

This kind of standardization.

A

That Gateway is intended to another big change. Gateway was the kind of like the addition, but basically just get the price from that pool or uh you can use their smart router.

A

We initially built Gateway, assuming that people would want to use a router, uh because we thought price would be the most important attribute to care about, but it turns out that the router um is a bit buggy and it's also a lot slower thanks to uh Community developer V hash off now you can't define whether you want to use a router or access a pool directly and by by default you're going to ask the support directly.

A

So, uh in this class release, uh we've been starting to add kind of more support for uh our gold and silver connectors. So we added candle, speed and Market order, support to ascendex. And finally, um we have a new connector order: Book decentralized, Exchange protocol called vertex. uh This is built on substrate, which is the polka dot ecosystem.

A

It kind of actually does not use Gateway, so it it connects directly by the client, which is actually a model that we think is worth exploring and finally, uh we're still doing regular meetings for our experimental, Pokemon dashboard Community project. We've actually made a lot of progress in the last couple weeks. On this, we sort of standardize the the base classes that every page should have, and so the next meeting you do, which I believe is uh next week, I think we'll have some pretty cool a few other things.

A

um So we've got a few more script examples, especially for directional strategies, uh there's also an amm data fee example, which I think will be really useful for people who are, you know, want to hedge something.

B

On a central exchange.

A

Based on something happening on a decks in one way, so overall we.

B

A

To encourage the community to both submit scripts uh and for us and merchand, because you know these scripts will be, should provide a library of examples that people can use but close-up for governance. I think we've really kind of solidified what governs means and what it's for, which is really to tell you how the community tell us, you know which exchanges should we actually invest our time and create bounties for and which ones are just not. You.

B

A

Being used so they should be removed from the third base. One last poll, these connectors didn't meet the minimum voting power threshold, uh so they have been removed from the code base uh and we've also removed the Arun officer strategy. We'll just do this poll every three months and goal is to have a set of really high quality connectors in the code base that are constantly evolving, because the exchange landscape and crypto keeps changing and what people want to use and what they want us to support.

A

What we believe will cost uh and finally, there's lots of other changes here. So so overall, I think that, from a development standpoint, homebots really kind of turned into a community-based code base, and we view our job as um not so much like writing a bunch of code, but kind of like you know, aligning the bounties and ensuring that development Tower is being spent, aligns with what personality actually wants.

A

But hopefully we can create the space for more and more community members to earn bounties uh and to actually do a lot of the work and we're yeah. That way, I think we can really leverage um the Tower of Open Source. So, with that um update on community bounties, uh the first one was this announcement of uscc bounties. uh We think it'll be better for people, because currently, the only really active market for h-spot is uni swap pretty high slippage.

A

uh Gas costs are like five dollars now, uh so there's a bit more process on our side, but we think this will be much better for uh the juicers for obviously, if.

B

You have any questions.

A

On this, you can contact Carlito, uh our community manager, who's managing this process on the support Channel. um So we have been seeing a bunch of interest in bounties.

A

So this is what our current Bounty score looks like, for instance, the ones we added last week were snapshots to approve to upgrade the mxc connector, as well as the coinbase spot connector, uh and then there's a few other kind of connectors ongoing and we've also been kind of just creating bugs bug bounties for simple bugs that, oh, you know, probably only take a developers, one or two hours to address.

B

But because of you know, there's lots of like.

A

Lots of little bugs in code base like this, and we think that you know because of context, switching required to fix these bugs I think they're, actually quite suitable for bounties, like that. Finally, we've also been kind of content bounties, so uh the the one that Drew man has is gonna show in a bit is actually one cotton Bounty. uh We have another one for the grid kind of grid strategy experiment. This is actually um was.

A

This was created by the same Tim developer, who created the community based a fixed grid strategy originally, and now he has turned that can so it scripts as well as he's created a video and explained uh how it works. uh So so I think these are kind of kind of like content bounties. We want to um create going forward because you know uh I think having the code as well as the explanation of how to use and customize it from the person who created the code.

A

I think is really helpful in having people understand how to run these spots. So if you're interested in kind of like creating more content bounties, uh uh you know, please let us know, and we can try to create more like that. Finally, I.

B

A

Caitlyn's can finally, but quite a lot of update you on. We have started to expand our community programs, so this is something that we started about um I think like six months ago, but every week we have been paying out 80 000 to hbot to um to basically you know to to a set of people on Discord proving answering questions so the way we do it is that uh you know every good answer on a public.

A

uh You know Discord Channel, um good answers might get either one a two or three Emoji, a penitent message by the foundation team uh depending on and that's really kind of like, depending on the quality of the answer. uh One is an answer that can answers. The question two uh refers to the answer that refers to um a.

B

Documentation or provide supporting.

A

uh Material that answers the question and three is- is a more in-depth support thread. In addition, support really isn't you know the more time we spend a support, the more time it takes away from writing code and creating bounties and everything else, we're doing, uh there's only seven people that work the Helena foundation, and so because of that we've decided to include us individually within this as well.

A

uh So in other words, but we, when we answer questions to Discord uh uh we're doing so in a personal capacity and you know performed by supporting users as opposed to um our normal capacity. That's why we're adding ourselves we're adding we're letting community members from um the foundation essentially uh take part in this effort. So the way this works is that is the fifth pool of h-bot, and uh you know this allocated for anyone that answers questions. It keeps on succeeding. We watch and allocate more HVAC.

A

We think this is a more scalable way of getting questions answered than having dedicated support uh Personnel, also because a lot of the questions answered are fairly Technical and involves specific, connectors or strategies. We want to try to find a way to allow people to have the expert case for audience uh to be able to answer questions and.

B

A

To be a reward for uh actually showing your expertise, lowers the bearer of entry for people to run a bot and feel like you know they have someone guiding them is actually really helpful, so we definitely want to create more and more uh okay. So um so that's really it for bounties yeah I wanted to invite on stage uh Drew man. Let me go over on the directional strategy. Theorem hi.

B

Mike, it's a pleasure to be here today, thanks for inviting to me, I will talk about this uh directional strategy. I have performed two weeks ago for, in the context of a community experiment,.

A

B

Is this energy is called the multi cumulative difference and I will start by explaining the the core behavior of this? This is a term follower. It basically relates on the belief that strength will continue, allowing traders to write the term for profits like this guy is making here uh we can identify these Trends with technical indicators and moving average, so I will be using uh the macd indicator that it's a pretty popular. So let me show you please this picture, so we can see it better here we have this popular MCD signal.

B

Is that when these lines crosses it's, it can be a bullish signal or a bearish signal. So look how much signals I have highlighted here with this little circles.

B

But if we look here, uh there is a kind of problem here, because these price movements are incorrected with this training signal, so we are getting kind of fake signals, so the idea of this strategy is basically trying to figure out all of these fake signals. uh If I had to choose two signals from here. I would choose this one, because it was a very good short position and this one, because it was a nice good position, so I like to make an analysis here. uh These are like C waves destroying signals.

B

So what we're looking for is to catch up waves like this and trying to ignore all of these ones that they won't take me anywhere. It was a follower based in this indicator, so.

A

We can say that.

B

When the histogram is positive, we have a bullish and when it's negative we have a bearish one. So the hypothesis by designing it for this experiment, is that the bug the histogram signal will be valid only when there is a sustain momentum change. What I'm trying to do here is to measure the direction holding of this of this histogram. So we looked at this very green bar I'm, starting to count here in a cumulative way all the differences between consecutive parts.

B

So what I'm doing here is trying to form exactly like acid wave I will be confirming this trend at this very point, because I have set up this threshold here that this threshold is basically trying to tell to my bot that write these Graves and not these ones. I'm trying to take this out. You can use pandas here, so you can do everything uh then you're going to get the signal here, I'm going to override the tape, profit and stop loss, size I will discuss this later and I will defining the long and short entries.

B

Finally, we're going to set up some class attributes like the.

A

Exchange trading bar.

B

Interval and so on, so I will start with this get processes. Data frame, method, uh I have prepared this table here. I, don't know if I would have enough time to talk about this, but I will jump directly into the method, because I think it's the most interesting thing. uh The first thing we're going to do here is we're going to love the candles uh here. I'm setting up this multi indicator.

B

Look that you only need to add the second line here: I'm using pandas TI. This is a technical indicators. Library here I'm defining this. So the goal is more sunrises, so I don't know. Maybe tomorrow, I want to run it with MRT compose it with around 15 30 and.

A

10 I just change.

B

These three numbers and that's okay, so I, don't need to change the code anymore. You start calculating this macd histogram normalizes I I. Invite you guys to.

A

B

The the video because I talk about normalization there so I can create in this volume. Then I am performing this. This difference, I'm just calculating this difference. uh Two minus one is one three minus two is one, so this is the difference between consecutive variables. This column is like an event tracker. Every time, I got a assign change here. I set this event like true so I can perform this cumulative Zoom here, uh giving an ID for every every group, so I can Group by and perform a cumulative 2 for every graph.

B

So this is what gives.

A

The indicator actually can I ask a quick question, so um so the negative tasks- slow and Signal numbers like 12, 26 and 9.. Yes, how did you uh come over those numbers.

B

uh They are default, uh they are the most popular configuration. Of course you can change in buyback testing, okay,.

A

So it's 12, is it like 12 or 12 days or uh like? Is it like.

B

Does it correspond.

A

To length or like it's.

B

Embarrassed so if you are doing that in it depends on your interval.

A

So so so I think I understand what's faster. What is Mac signal.

B

Theory, but you are just basically comparing to two moving average, so these first two are uh are calculating this difference. uh I.

A

B

Really want know how to explain this, because uh I I can read in Spanish, so in English.

A

The macd I think um I think a rough understanding you're getting like a um one line which is like the trend for the last 12 bars. That's it fast.

B

You're getting another line.

A

Which is a trend or amazingly.

B

A

Which is the 26th uh bar signal.

B

Yeah, so that's right.

A

Let me see signal is that also.

A

Something else.

B

Yes, we did 12 and 26 values, you are calculating a line and then, with the other one with the nine you are comparing it. So you can see if the the trend is comparison or the operation Zone as well. It's a little bit complexity. It uses exponential moving average information.

A

B

They will certainly do it better than me. English isn't my even my native language, so.

A

No, no yeah, your English is a lot better than Spanish.

B

Okay, so I will move on to the get signal method uh here. Look that I am presenting these two initial conditions at the very same bullish signals uh and I'm, adding these conditions so with.

A

The Magnificent.

B

Thresholds I mean this is like the anti-waves condition.

A

B

What it's doing it's very simple: it's getting this processing data frame that I have just show you and I'm saying these two thresholds here and again in the last row. So I can perform this very simple logic. So I can get this signal. uh One is for wrong. Minus one is for short and zero is nothing so doing this I will jump into the the strategy config.

B

We need to choose the site because we can play both a long and shorten the positions and.

A

We can edit them.

B

By the graphics, stop loss or time limits, these are like the entire universe of results you can get, but I like to talk about the seventh scenario. That is this one I, don't know if you watch your robots performing live, but if you do and you're at this point, you will probably get like this guy, because you you were really so close to to reach to the Supreme barrier. uh You finally ended by exiting the Surplus position so to do this. uh This is the directional strategy based class implemented a very cool feature.

B

That is, the training, stop loss. So at this case, if the price moves enough to activate.

A

B

Train activation, it will appear this training Delta bar. This is just moving. This stop loss barrier up here. So in the case you are at this stage and you will just exit here with less profit, but you won't lose money. So this is a very good feature. Yeah.

A

That graph is actually part of the this chart is actually.

B

A

Best explanation for why you should use.

B

Okay, so now uh I I have also talked about this in the video These are like Dynamic types of those positions uh I'm using this time environment, because uh I have uh written this book and advances in financial machine learning this measure to even perform the triple Bar Method, so I we have basically with fairy and my friends and.

A

B

Here to do this and you you just calculate your standard deviation from the close price. You normalize it and then you can set up these significance of blocks. I don't know. Maybe if, uh if you set these multipliers to one, you will get one standard deviation up at one down, uh you can use 0.5 or 2, maybe I don't know, but the you you need to to adapt to the market, because the price moves continuously and.

A

B

You set a constant position, then Market changes. So it's a good idea to do this. This is another practice in parameter to the train.

B

uh I think the target variable in the previous code, it's very simple in get processing data frame is just pandas. It's very simple to do this. You just calculate.

A

B

Activation and divide it by the growth it's a vectorized operation.

A

B

There is no for Loops anything and in that signal method you just Define this threshold. uh This threshold is important. I have mentioned, because uh here I'm highlighting this vessel.

B

Is a 0.06 percent, so if the market is to provide, you need to set up this threshold, because if you are trading, if it values right, then even that if you make a take profit, you will pay fees for that and you will lose money. So it's uh it's necessary to set up this so I'm here, I'm overriding these State graphics and stop loss positions in this. They are class.

A

B

So it's reasonable! You just need to put this Target and multiply it, but like this multipliers, so it's very simple to implement this. So these are the class attributes I mentioned before here: we're defining uh the name, the search name, the training pair The, Exchange interval order amount whatever and.

A

B

Defining this position, configs.

A

B

Limit that is in minutes of activation data and our trading Delta.

A

B

Nice so now it's time to deploy I have prepared this for you today, I will open my terminal.

B

I am using Ubuntu and it's very unfriendly for this yeah but I I want you to show you the about performing live, so you can see the candles feel and you can see the multi-cumulative indicator on live and yeah. Maybe if you're lucky, we can see a position opening okay.

A

B

Anyway, I will jump for this now. I won't lose your time. I'd like to share some of the results. I got here. They they went pretty well. In fact, I lost money. I didn't want to sell this up to this moment, but the the good thing uh even of this is that it's a very balanced uh about I mean sales and buys orders where.

B

Feature so it worked in some way, so I'm happy for this uh I'm. Finally, uh remember that I was saying that I would I wanted to write time waves. So I think that this is happening.

A

Here so even I.

B

Have this money I think I have validated this idea because.

A

Here we can see.

B

That in this service markets they are very easy, very movement and I think. Why is this happening here? But at this point at the big words uh I'm surfing uh now, I I need to teach that server that he needs to jump out of the way in a certain moment,.

A

I need to think in a complementary.

B

Indicator such as I know maybe RSI, or something like that. That tell me: okay stop jump, so he he doesn't crash so.

A

B

B

Any of the settings before you do this or.

A

Do you just run it run the the same.

B

A

You had you showed it before before in.

B

Fact, yes, uh that's a very good question because uh during this experiment in live performing uh needs to be realized that most of the attack testing tools are based in the close price. So it depends on your resolution, of which candles can you turned out uh to get results and that's different in.

A

The in the reality.

B

Because look at this hero testing would only have given me uh one two, three four positions and look: how many, how I open here so thinking about the testing yeah in a high resolution, way I. Think it's a better solution to do this uh or you can just run your Bot I mean.

A

B

This is 10 second.

A

Candles right, 10 seconds.

B

uh Yes, but I say performing several.

A

B

uh Maybe every time you add a new variable uh so using one second fantasy is really prohibited.

A

But overall, though, I think actually I, actually your result is actually a pretty good one which is like you know. It's.

B

um You're, usually really.

A

Only using kind of like very simple indicator uh and.

B

A

Able to get 46, you know um yeah.

B

Yeah, so they should be able to build.

A

On top of that right, but I think this is.

B

Actually, I would think something that a.

A

Lot of new cargo Traders don't realize, which is it's like? It's not like a binary make money or.

B

A

Money thing: it's.

B

A

Like you know, you're, you know, hopefully your starting point is you're not losing that much usually.

B

You're always losing.

A

Money like it was.

B

Easy it was that.

A

Easy to make money because everyone will be doing it so.

B

Initially, it's more like.

A

You want to try to find um a strategy that has potential and then.

B

A

Iterate on that strategy and hopefully be that strategic Point always to be profitable, like 5100 win ratio and then.

B

A

When you do that, then you just keep on running that strategy, but you should expect that eventually it's gonna, you know the window is gonna.

B

A

B

Think it's more like an iterative.

A

Process like a it's a special probability.

B

A

Rather than like, you know, yeah.

B

Exactly uh another thing: I I like to hear is that uh this community difference concept you you can use it in other stuff, apart from the MLP, because, for example, when you watch volunteer events, they can be diverging and when they are talking about, you can measure it with this kind of indicators. So it's a very simple way of avoiding I, don't know, maybe the relative functions or calculating intersection handles or for these uh trying to measure this intensity. So it's a very cool tool.

A

Is adding a total like how big the difference is.

B

A

It's like a direction holding measure.

B

That's it essentially, nice.

A

All right, so one comment from Bojack. uh He says.

B

um Yeah, like He, suggests.

A

That you had one factor to the strategy like order size um and uh yeah. So is that something you've uh thought about or.

B

Considered, uh yes, I think that if we are using something like a machine learning model, this is like this. You can use the predict probability as a size measure uh I've been working with trading a lot and he does a kind of a guardian.

A

Stuff so I'm trying.

B

To learn from him.

A

That's definitely I think where uh yeah I think once we have a channel Speed, uh we have the the practicing module. You know down the dashboard and people who really start to do. You know true true machine learning and, and maybe.

B

Even like some deep learning you.

A

Know give them like okay, something else. It's real AI in here.

B

Yes, I think that we, the humans, can can barely understand what's happening, but I I think we should delegate that to to the machines they are better than us in that way. Yeah. Definitely so this I'm reading the chat. If there is any other question.

A

B

So yeah actually there's one.

A

Question for more Jack, which is yeah, is.

B

This a lagging signal and if so.

A

How far is the lag behind current price action.

B

Oh, that's a very good question. uh Let me think when you mean lagging.

A

B

A

B

Your treasure I mean, if you set up I, don't know if I answering, but you are asking. But if you choose a little filter, then the last will be lower and if you choose a very high threshold, you will you light so much. But you need to be careful with this because you, you want to be very late, but you don't want to be very low. I might think in some some Dynamics threshold so that.

A

Would be nice, yeah yeah? The way I would answer, questions I think like the goal is always to try to get a bit easier. You know because, like.

B

You want to try.

A

To predict where class is going to go.

B

um But I think there is like an R.

A

To it, because it's like, if you just kind of follow the price, then then you can be trade too often and not really making money, um oh so yeah! Ideally you only want to like trade. You know like you want to call it. The balance between, like.

B

Okay: okay, yes, another way you can see, this is I, didn't think about this lagging scenario. uh I won't give a fast answer because I don't know the answer. Yet uh uh this can be useful uh to fit. If you are doing mean regression, we can perform very nice middle version strategies uh and every time we lose money is when there is a very high price movement. So you can use this as a filter too uh I say you can enter here. Okay, you can use it to don't enter there.

B

Well, that will be another feature if you.

A

Want to yeah um yeah.

B

So so, but all.

A

Right, thank you, German, for, for uh you know, preparing this and doing all this work so uh and I think this is really um a great example of what.

B

A

To enable uh you know, it's now more like a framework where people can create any type of strategy. Dave wants, uh so so we want to kind of encourage people to create. You know whatever they want to create and.

B

A

The space and these kind of.

B

A

To uh you know showcase what they built, and so we can all learn from each other.

B

Thank you, Mike uh I'm really happy for doing this, because I've been doing this for several months, and this was like the first opportunity. I have to make something.

A

Real and show it to other people, so thank you.

B

For for making this possible.

A

Right well, yeah, anytime, so yeah we'll do a second. Yes, if anyone else wants to create their strategies and create content, please let us know we definitely want to encourage it.

B

A

Thanks everyone I.

B

A

Next month,.