I Built a Sports Betting Bot with ChatGPT

Summary

本视频由Siraj主持，展示了如何利用Chat GPT和一个名为GPT Wager的应用来构建一个体育博彩机器人，并通过这个机器人对金州勇士队和布鲁克林篮网队进行了两次共计2000美元的赌注。视频深入讲解了从简单的套利机器人开始，逐步升级到使用XG Boost和深度学习模型，再加上Twitter情绪分析来提高体育博彩模型的准确率。Siraj强调了数学在制定盈利策略中的重要性，并通过实际操作演示了如何搜索和使用开源代码和API来实现这一目标。最终，机器人成功赢得了7000美元，展示了AI在体育博彩领域的潜力。

Takeaways

• 🤖 Siraj构建了一个使用ChatGPT的体育博彩机器人，该机器人基于数学预测和不同技术的集成，包括套利博彩、XGBoost和深度学习。
• 📈 强调了数学在体育博彩中的重要性，尤其是套利博彩、期望值博彩和其他数学技术的应用。
• 🔍 介绍了套利博彩的概念，即利用不同博彩公司对同一赛果的赔率差异来保证盈利的策略。
• 🐍 展示了如何使用Python和开源API（如Odds API）来获取NBA数据，并以此为基础构建套利机器人。
• 📚 通过Github搜索和现有的开源项目，减少了从头开始开发的需要，展示了重用现有代码的价值。
• 💡 引入了使用机器学习模型（如XGBoost和深度学习）来预测NBA比赛结果的进阶方法。
• 📊 结合了Twitter情绪分析，使用公开的Twitter数据和文本分析技术来增强预测模型。
• 🌐 展示了如何将机器人集成到Web应用中，并使用web3和加密货币进行博彩。
• 💰 机器人成功地从两次投注中赚取了7000美元，证明了使用技术和数据驱动的方法在体育博彩中的潜力。
• 👨‍💻 Siraj鼓励观众订阅和点赞，表明他计划继续制作类似的教育内容。

Q & A

Siraj在视频中创建了什么类型的机器人？

Siraj在视频中创建了一个体育博彩机器人，使用了Chat GPT来进行体育博彩预测。

Siraj的机器人使用了哪些数学技巧来进行体育博彩？

Siraj的机器人使用了多种数学技巧，包括套利(Arbitrage)、XG Boost和深度学习，以及在推特上进行情感分析来改善体育博彩模型。

Siraj如何验证他的博彩机器人是否有效？

Siraj通过在两个队伍上下注（金州勇士队和布鲁克林篮网队）并展示他的赌注结果来验证机器人的有效性，最后他展示了机器人帮他赢得了多少钱。

Siraj的机器人在预测体育博彩时使用了哪些数据源？

Siraj的机器人使用了多个数据源，包括不同的体育博彩书籍(即不同的博彩公司提供的赔率)和推特上关于球队的情感分析。

视频中提到了哪种特定的博彩策略？

视频中提到了套利(Arbitrage)博彩策略，这是一种通过在多个博彩公司上对所有可能的结果下注来保证利润的策略。

Siraj是如何使用OpenAI和其他工具来构建他的博彩机器人的？

Siraj通过使用Chat GPT来生成代码和策略，利用Python和各种库（如Keras、scikit-learn和TweePy）来处理数据和模型训练，以及利用Google Colab和GitHub来运行和查找代码。

视频中提到了使用哪个API来获取NBA数据？

视频中提到了使用Odds API来获取NBA数据。

Siraj如何整合推特情感分析到他的博彩模型中？

Siraj通过使用TweePy库来抓取有关NBA球队的推特数据，然后使用TextBlob库来进行情感分析，分析人们对球队的正面或负面看法。

Siraj在视频最后展示了什么结果？

在视频最后，Siraj展示了他的博彩机器人在两次下注后赢得了大约七千美元的结果。

Siraj提到了哪些编程实践来简化他的博彩机器人项目？

Siraj提到了使用单个类文件来避免大型项目和多个依赖关系，以及利用现有的GitHub资源来简化开发过程。

Outlines

00:00

🤖 构建体育博彩机器人

视频介绍了Siraj如何使用Chat GPT和一个名为GPT Wager的应用来构建一个体育博彩机器人，并展示了他在Golden State Warriors和Brooklyn Nets上下注的两笔赌注。视频的主要目的是展示如何利用数学和不同的博彩策略（如套利、XG Boost和深度学习模型）来构建这个机器人，并通过在Twitter上进行情感分析来增强模型的预测能力。Siraj强调了数学在赚钱过程中的重要性，并计划展示如何从头开始构建这个机器人，包括如何利用Chat GPT询问最常见的数学技术和如何实现一个简单的套利博彩机器人。

05:00

🔍 寻找开源赔率API和构建简易套利机器人

继续探索如何构建体育博彩机器人，Siraj讨论了寻找合适的开源赔率API的重要性，并选择了Odds API作为数据源。他展示了如何使用Google Colab运行Python代码，以及如何处理遇到的API问题。此外，他强调了在不同的体育博彩书上寻找套利机会的概念，并尝试创建一个简单的套利博彩机器人。通过实践，Siraj揭示了在构建这类机器人时需要考虑的一些技术和挑战，如数据清洗和处理。

10:01

📈 利用GitHub资源加速开发

在体育博彩机器人的开发过程中，Siraj发现了一个GitHub上的项目，该项目已经实现了他所需的功能。这表明，通过有效利用现有的开源资源，可以大大加速开发过程。他进一步探讨了如何使用这个GitHub项目来获取和处理数据，以发现套利机会。通过这个过程，Siraj展示了如何利用现成的代码和API来简化开发工作，同时也指出了在整合和使用这些资源时可能遇到的一些挑战。

15:02

🐍 构建深度学习和情感分析模型

为了提高机器人的预测准确性，Siraj探讨了如何整合深度学习模型和Twitter情感分析。他提到了使用Tweepy和TextBlob库来分析推特上关于特定NBA球队的情绪，并考虑了如何创建一个深度学习模型来预测赢家。此外，他讨论了利用GitHub搜索功能找到现成的NBA机器学习模型的价值，这再次强调了利用现有资源来加速开发过程的重要性。

20:02

💻 集成模型到Web应用和盈利结果

在视频的最后，Siraj展示了如何将他的体育博彩机器人集成到一个Web应用中，并通过使用一个去中心化的博彩服务Dexport.io来实现赌注。他使用React和Firebase创建了前端，并通过MetaMask和Polygon网络来处理交易。最终，他揭示了机器人在Golden State Warriors和Brooklyn Nets的赌注上赚取了七千美元的利润，证明了这个项目的成功。此外，他鼓励观众订阅和点赞，以支持他继续制作类似的教育内容。

Keywords

💡Sports betting

The main theme of the video is using AI and data science techniques to build a sports betting bot.

💡Arbitrage betting

A method to exploit inefficiencies in betting odds across sports books to make low risk guaranteed profit.

💡Sentiment analysis

Analyzing positive/negative sentiment on social media to predict team performance.

💡Machine learning

Using ML models like neural networks to make betting predictions based on sports data.

💡Computer vision

Analyzing past game videos automatically using computer vision techniques.

💡Cryptocurrency

Using a decentralized crypto sports betting service to place bets.

💡Github

Searching Github for existing code to reuse instead of building from scratch.

💡API

Using sports data and odds APIs to get input data for models.

💡Web app

Building a web app to showcase sports betting predictions and bets.

💡Math/statistics

Core techniques like probability and statistics that enable sports betting algorithms.

Highlights

使用套利投注机器人赚取少量利润

利用概率和统计学构建机器学习模型

结合Twitter情绪分析改进模型

Transcripts

hello world it's siraj and I built a

sports betting bot with chat GPT and in

this app called GPT wager you can see

that I've made two bets the first one is

about a thousand US dollars worth on the

Golden State Warriors the second one is

about a thousand dollars on the Brooklyn

Nets and this is because of my bot it's

because of the predictions that it

output and in this video I'm going to

show you how I built this bot what the

results are at the end you'll see if I

lost two thousand dollars or I made up

to as you can see a combined total of

about ten thousand dollars that's at the

end of the video so stay tuned but

before we get there let's build it

together with chat GPT let me show you

how I built it and we're the most

important part of this video and what I

really want to stress to you is

mathematics and how awesome mathematics

is mathematics helps you make money and

in this video we're going to start with

a very simple bot it's an Arbitrage bot

I'll explain what that is then we'll

improve it to be what's called an XG

boost spot we'll improve it again to be

a deep learning bot then we'll add deep

learning plus sentiment analysis on

Twitter so we can see what people are

saying on Twitter about a team and use

that to improve our sports betting model

all right so that's what we're going to

do in this video so the first step for

us as we build this with chat GPT the

first step for us is going to be to ask

it a question that question is going to

be show me a list of the top 10 most

common math techniques now remember we

love maths we're going to ask you about

math techniques and we're going to be

very direct to make money from sports

betting you know I've heard terms like

Arbitrage and let's let's give it some

context here remember chat GPT remembers

context so I've heard terms like

Arbitrage and expected value betting I'm

not sure if that's helpful or related

just like we would talk to a human

period let's just see if it's going to

help us now I have to re-log in because

it's been a while that's always I know

annoying to have to do that let me paste

that back in here

very real stuff that's how it

that's how it is with chatgpt so it's

going to list a bunch of different

techniques and all of these techniques

are going to be in the categories of

math like probability that's about

likelihood and whether or not something

will happen that's what it's concerned

with then we have statistics and that's

about empirical data usage it's a

collection of tools to analyze data and

then we have algebra arithmetic right

numbers plus minus subtraction Division

and even calculus in the case of Markov

chain Monte Carlo that's a way of

simulating different outcomes and we can

use calculus to find the rate of change

or the derivative of different variables

so we can see 10 methods right off the

bat that chat GPT gave us to make money

from sports betting and we don't know

what any of these terms are yet because

we're noobs I mean we know a few terms

but we're going to pick one of them

we're just going to pick number two

Arbitrage betting and that's going to be

the first one that we're going to pick

and what is arbitrage betting Okay so

Arbitrage is this idea of in the sports

betting space we have all of these

different sports books and sports books

are always betting on the odds of

different results whether one team wins

or one team loses whether a certain

player is going to do well or not all

these things are odds and they use

quantitative models to predict these

odds and they're really good at this

what Arbitrage betting is is it saying

that hey if I bet on the on all possible

outcomes across a variety of sports

books because they all have different

odds for the same outcomes I can find

these inefficiencies in this market

because it's very similar to a financial

like a stock market it's like a sports

betting Market I can find very similar

inefficiencies and then I can exploit

them to make money and so the if the sum

of the inverse of all of the

probabilities of the odds of a given

game are less than one we can say that

an Arbitrage opportunity exists so even

if we make two bets in two different

directions

if there is a real Arbitrage opportunity

we can be guaranteed a return but that

return isn't going to be that big it's

going to be between one to ten percent

max and the sports folks are going to

get to get wind of what we're trying to

do and they're probably going to ban us

so it's not the best technique probably

but it's a good place to start so let's

ask it to do that let's

um ask it to

build us a simple Arbitrage bot so show

me an example

of an Arbitrage bot

in Python for sports betting and have it

be real simple have it be super simple

and fit into a single class file because

we don't like giant projects with many

dependencies and we're going to be very

specific we're going to say it uses

mathematics to output and we're going to

be very bold as well a provably

profitable strategy okay and then we're

going to be very

needy with it then explain the math

behind it to me

okay

and hopefully it gives us a working

example and it did before I swear to you

but right now it's decided that this is

um not what it wants to do but

it might let's wait for chat GPT please

chat GPT do this for us we need this to

happen

um it can give us an example thank you

ah show me a python Arbitrage bot for

sports betting simple example that fits

into one class file I think adding the

math thing it didn't like that so given

that using a single library and let's

say three different book markers we can

do you know a three-way Arbitrage as

well it's going to find that Arbitrage

so let's take this code and let's go to

a Google collab notebook

collab.research.google.com we'll open

that notebook it's just an easy way to

run python code even if you're not like

a super good code or anything so we'll

paste that right in there and we'll run

that and we can see that there's already

an issue with this and the issue is that

this

api.bookmarker1.com is not legit so

we've got to get some legit Sports data

so let's ask it for some of that so

that's going to be our next question

going back to our original prompt series

here and we're going to say show me

let's go back here show me a list

of the top 10 open source odds apis for

sports betting we don't just want we

want several

and I have gone through some of these

and it's it can be quite a pain to find

a developer API given whatever area of

the world you're living in so

um in the end the one that I found that

would work is the odds API and here it

is odds API right here so that's the one

we're going to use the odds API use the

odds API

uh in Python to pull NBA data

pull NBA data

and then it's going to give that to us

and then what we're going to do is we're

going to sign up for the odds API

because we need that and here it is odds

API we can see we need to get an API key

it's going to start out free perfect

enter our name and everything assume

we've signed up for that and once we've

signed up for that we're going to go

back to

the main page and we're going to get

that API key where is it

it's under account here's our API key

okay and what we can do is we can go to

sheets and it's got this Google Sheets

integration where it can just pull that

from our Google sheet so let's start

with that we'll start with a simple

Google sheet we'll make it a new Google

sheet now we've assume we've installed

this add-on which I've already done and

once we have that add-on we can go to

extensions sports odds start and it's

going to pull up a live odds API we can

paste in our API key that it gave us

back here under account

and then we're going to populate this

Excel spreadsheet with all of the NBA

sports we're going to pick NBA from the

list here basketball and then we want it

to be in decimal versus American

and then we'll fetch it okay so here it

is

we've got latest the latest data right

here from different bookmarkers like

bookmakers like DraftKings and Bovada

and all this stuff and what we can do is

we can find the we can Arbitrage the

odds and basically compute what that

profit is going to be so using this odds

API I wonder if they have some simple

python examples for us hopefully they do

so we'll go to home we'll go to code

samples and then boom they've got some

python examples here running on replit

we'll show the files it's going to be a

main.pi file we'll just take the entire

thing here

and we'll copy it go back to our

code here

and that compiled it and now we have

those in python as well so now what we

have to do is we have to compile this

data that we pulled using the odds API

with that Arbitrage bot that chadbt give

has given us and that's going to require

data cleaning we're gonna have to take

that API clean the data and then process

it insert it into these two definitions

this is going to take some time and

energy now let's we can do that but

before we do that let's just do

something really quickly just to make

sure that we save enough time let's just

go to GitHub real quick and we're just

going to search a single search term

that's all we're just going to search

for statistical

Arbitrage

for sports betting just to make sure

nobody's done this be oh there's one

right there by Ryan crewman's knocker

thank you Ryan for this and it was made

four months ago very cool it's using the

odds API okay perfect uh that's exactly

what we need to do what Arbitrage okay

we didn't even have to do any of this

work this guy's already done it for us

and that is the value of getting good at

searching for code on GitHub because

there's so much value to be found there

so let's run this thing this guy's got

an IPython notebook for us and it's

going to create an Excel spreadsheet

just like we found before the odds API

it's going to get all that and then wow

that's a lot of data once it's got that

data what's going to happen next look at

all of this parsing that it's going to

do find the number of possible outcomes

find the best odds determine the odds

and then you know extract the each

individual bookmaker we would have to

write all these functions ourself we

don't want to do that so let's go back

here and we're going to download this

and upload this to Google collab so

first step go to download zip we clicked

on download save the file it's saved now

we open the zip and we're going to

upload it to Google collab so we'll go

to colab.research.google we'll go to

upload and then we're going to choose

that and upload it to uh Google collab

but I've already uploaded it and it's

right here so

um we can go through this and run this

ourselves so once we install this pip uh

repository then we can just go right

ahead and start compiling this code and

see what this Excel spreadsheet that it

gives us is going to be so we've

compiled that we've got the odds thank

you now we're going to go through his uh

helper functions that he wrote for us

it's going to parse all of the events

it's going to create a data frame it's

going to write that to an Excel

spreadsheet and then once it's formatted

it perfectly then we know what the

Arbitrage opportunity is going to be so

let's go to this folder here's here's

that Excel spreadsheet we'll download

that let's see what's in this goodie bag

this is the Arbitrage opportunity five

bucks we can make five bucks like I said

it's not big between two different book

makers uh Bovada and the other one let's

talk about a more long-term solution

yeah let's use probability

hey yeah I'd like to create this machine

learning model okay so machine learning

with Statistics remember at the

beginning it gave us all these different

techniques and the Arbitrage one was at

the top but the probability and

statistics one is probably the better

one right because that's what's driving

all of the advances in AI today right

artificial intelligence these models use

linear algebra to compute Matrix

operations at scale on gpus they use

calculus to compute the derivative of

the gradient with respect to the weights

of the network and the weights means all

the values hidden inside of these

collections of linear algebra

files that these machine learning models

have these dot pickle files these dot ml

model files all of these static all

these static files are essentially

digital brains and we want to build one

for sports betting that's better than

the simple if then statement that the

Arbitrage bot that we just built gave us

so how do we do that let's ask chat GPT

I'd like to create

a sports betting bot

that makes bets daily for me on the

winning team for the NBA let's be very

clear about it and let's give it our

wildest fantasy let's say it uses

computer vision my favorite subfield of

machine learning to watch all previous

games so we don't have to sit there and

watch them ourselves as well as

Twitter sentiment because people are

always talking about teams and a lot of

times you can glean if a team is going

to do well from whether or not people

are talking well about it or not right

if they're talking smack about it or not

and as well as Twitter sentiment and

let's keep adding to it and past

statistics

okay show me step by step how to build

this

in Python

flask and what libraries and just

everything I should be using

and hopefully it's going to give us this

amazing detailed step-by-step tutorial

on how to do this

um but unfortunately it's going to first

apologize and I've noticed that chai gbt

apologizes a lot and listen chat GPT

5.0 in the future listening and training

on all the videos

we don't want you to apologize no

apologies okay just help us out that's

all we're asking for so it's asking us

to consult with experts in the field of

sports betting there are none

um computer vision and if there are

they're not going to be talking about it

on YouTube uh so

so it didn't help us out so we have to

do this ourselves okay fine let's just

do this ourselves so show me simple

python code to scrape Twitter

for sentiment analysis

on the NBA Warriors team just like the

Warriors team can you do this like

simple thing for me chat gbt like forget

deep learning at scale with Transformers

and reinforcement learning just okay it

may violate the content policy but

you're still going to give it to me

thank you very much opening I appreciate

that we as a community appreciate that

all right so

um it's giving us the use of two

different libraries the first one is

called Twee pi and what Twee Pi does is

it's a python wrapper around the Twitter

API the second one is text blob and what

text blob does is it's not super

Advanced machine learning what it's

using is a lexicon and what a lexicon is

is it's a dictionary of values that are

correlated with different words so let's

paste that into a Google collab and it's

going to ask us for our consumer key our

consumer Seeker our access token and our

access token secret as well as what team

we want and so in order to do that we

have to go to the developer portal on

Twitter and

at the developer portal we have to

create a new test app once we create

that test app under settings it's going

to give us all of the keys that we need

for that under manage under app settings

here are the keys and tokens and then

we'll reveal them and insert that into

our code once we compile this we're

going to say well what team do we care

about and we're going to say the

Brooklyn Nets and already it's given us

the list of positive tweets and negative

tweets and we can construct a very

simple model based on just this we can

say if there are more positive than

negative tweets don't make a bet else

make the bet because the team will win

because people really love this team

right now you know that's one very

simple model but let's keep improving it

so instead of just doing this Twitter

sentiment analysis bot let's now add

deep learning to the mix okay so let's

ask it that question so let's say

um chat GPT

uh show me a deep learning model to

predict the winning team

given Sports data

just something simple like

one sentence maybe it's going to do it

this time I hope

thank you okay so what it's probably

going to do is use the Keras library to

construct a neural network the easiest

way to construct a neural network thanks

Francois Chalet

um and scikit-learn to build that model

this is going to be a neural network

very simple stuff watch my videos a lot

of neural network videos and it's going

to train it on the CSV file that doesn't

exist we got to connect that to the

sports data API and then we're going to

have to go through the very tedious

process of training this model

on all of this Sports data and that's

going to take some time so we're going

to train this model we're gonna have to

do some feature processing what are the

features we want in the model and the

number of wins the number of wins what

are the statistics how much data then

we're going to have to run this at scale

and that's going to take some time so

let's do that we can ask chatgpt all of

these questions

what features should I encode what what

should the training testing split be is

my model good enough here are the values

we can give it the outputs and all this

stuff but before we do that once again

let's just go to GitHub and search and

I'm just going to do a very simple

search just three words NBA machine

learning and let's see if it gives us

anything and lo and behold Kyle scom has

already made an NBA machine learning

sports betting

system I don't know if it's a bot

because it's not actually making the

bets but it is using tensorflow and XG

boost respectively to create two

different sets of predictions right

it's using a neural network with

tensorflow to predict the winning team

and then it's also using an XG boost

algorithm to predict the winning team

and then you can compare both of them so

what we can do is we can combine our we

can combine several things we can

combine Kyle's model here with our

Twitter Twitter sentiment analysis model

we can say if Kyle's model predicts a

winning team and our Twitter sentiment

model says that this is going to be a

very positive sentiment winning team

then we can bet on the winning team

right and what we're going to do is this

is going to give us a lot of numbers so

we're going to summarize all those

numbers with gpt3 Okay so let's take

this model that Kyle has and we're going

to run it in a Google collab so we're

going to take this copy it so what this

is going to do is it's going to clone

that repository into the cloud it's

going to install all the

requirements.txt and it's going to take

this pre-trained model and what Kyle did

is he trained this model on the past

decade of NBA games and you can see many

many many Rose many columns what are all

of these columns what are all of these

acronyms I'm sure some of you sports

fanatics know mention it in the comments

I have no idea there's a lot of them but

that's the model they use to train on

all right up till today given the odds

from uh given sports book in this case

we're going to say FanDuel

it's going to predict given two

different models both the XG boost model

and the neural network model what the

expected value for each team is going to

be what is the expected value it is just

the likelihood that they're going to win

and we can see that the expected value

is

going to be pretty high for the New

Orleans Pelicans and the Golden State

Warriors and that's according to the XG

boost model but in the neural network

model

it looks it actually looks very similar

so that's the first part then we can

augment that with tweets then after we

do that then we can install the openai

library to then summarize all of that

okay the winners are here are the teams

and here are the losers much cleaner

much better okay here's the last part

how did I fit it into this web app so

what I did was I took my react startup

template and it's just integrated with

Firebase and then I added decksports.io

to that and dexport.io is a

decentralized web3 betting service and

it's the one that I'm using because it's

decentralized it's uses crypto and

anybody can do it anywhere in the world

which is really cool and once you sign

in and the way to sign in is using a

wallet I've signed in with my wallet

which is uh metamask and once we sign in

it's going to ask what network we want

to use I'm going to sign it and then

it's like well what network I'm going to

select the polygon Network and then I'm

going to use usdt which is USD tether

and given that I'm going to go to web3

sports betting and I I framed it into my

original

web up here so I could see the results

of my predictions as well as making bets

and you can see here the two unsettled

bets that I made with my wallet let's

see if I made money or lost money

all right it's the day after in drum

roll please it looks like the bot made

seven thousand dollars from two bets one

for the Warriors and one for the Nets

thank you AI all right thank you guys so

much for watching

um I want to keep making videos like

this every single week so if you want to

keep watching Please Subscribe that's

what really motivates me to continually

do this and like the video as well that

helps promote it for now I've got to go

find the optimal prompt so thanks for

watching

foreign

[Music]