7 Super Simple Steps From Idea To Successful Data Science Project
Ever had this great idea for a data science project or business? In the end you did not do it because you did not know how to make it a success? Today I am going to show you how to do it.
By Andreas Kretz, Data Science & Big Data Adventurer.
Ever had this great idea for a data science project or business? In the end you did not do it because you did not know how to make it a success?
Today I am going to show you how to do it.
To keep this article as real as possible let's talk about a distinct use case.
The theory is that you can use twitter data to identify current big data trends.
The goal is to to sell this service to customers for a topic of their choice. We are going to use AWS services to realise all of this.
I am not advertising for AWS, I just have more experience with it. I am sure Google Cloud or Azure have the same features, only named differently.
Step 1 - Get the data
Before you do anything else, you are going to need to prove your theory. To do that, you have to collect some data.
Twitter data can be accessed through public APIs. All you need to do is write a small program that can download tweets from certain users.
Use the programming language you are most familiar with. I for instance like Java a lot, so I use Java for these tasks.
When your software is working, download the tweets from all the influencers in the selected field. Influencers are people that have a large following and that tweet a lot of big data stuff.
For big data it is particularly easy because KDnuggets has a list of Twitter influencers: www.kdnuggets.com/2016/02/big-data-top-influencers-brands.html
To do it in the cloud, you can spin up a simple AWS EC2 Linux instance (nano or micro), and run your software on in.
OK, but how should I store the data?
The best way to store the data is to use a simple .csv format. One line per tweet that includes the tweet's text and meta information.
The meta information you should include is person, time, replies, retweets and likes.
When you are finished, upload the file to S3.
How much data should I extract?
My advice is always to get as much data as possible in a reasonable time. Let your program run a few days.
Twitter has some strict API rules for how much data you can query in a certain time. You have to throttle your software to not go all out. It will get timed out otherwise.
Anyway, a few months of tweet history should be enough. How much data to get is not an exact science, follow your gut.
One more thing: Only collect as much data as the machine you are using for analytics can handle.
Step 2 - Select the right tools for the analytics
After getting the data you need to select the proper tool to analyse it. Write down a list of analytics features you think you need and compare available tools.
Cheap and quick is the way to go.
You could either go user friendly with graphical tools like Orange, Rapid Miner or Knime.
If they don't really fit then go the write the analysis yourself. Python and R are amazing languages for data science.
But I love using Matlab, can I use it?
Of course, if Matlab has the features you need - use it! Use what ever fits your needs best.
You can use the previous EC2 instance to do the analytics on. Or stop the old one and spin up a new one if you need to change the operating system.
Step 3 - Prove your theory with science
You have the data and the tools in place. This means you are ready.
It's time to work the data and prove your theory!
Start by pinpointing trends you already know are in the data. A simple way would be to search google for significant events that have been reported much.
Try to create an analytics process that finds these trends.
How do I know when the theory is proved and it's time to move on?
If analytics can find the trends you specified, then you are on the right track. Look for instances where analytics finds new trends.
Confirm these trends, for instance by searching the internet. The results are not going to be reliable 100% of the time.
Before continuing you need to decide how much falsely reported trends (the error rate) you want to tolerate.
By the way 0% error rate is absolutely unrealistic.
Step 4 - Figure out your business model
After you got the science right, you should take a step back. Before continuing, you need to figure out your business model.
Ask yourself: What is it that you do, what resources do you need, and what value do you provide to the customer?
Who are your customers and how are you going to sell your product to them?
For what values are customers going to pay?
A nice way to do this is the business model canvas. It's simple and cheap, you can basically create it on a sheet of paper.
When you are finished, continue with building a minimum viable product (MVP).
Step 5 - Build a minimum viable product
After proving your theory, it is time to start building a first version called minimum viable product (MVP). The goal of a MVP is to build a solution that only delivers the core functionality.
Don't go for fancy solutions. Focus on the main functions you need to realise.
Stick to what you know and what will work in the beginning and expand your system later. It can be something really simple like a RDS database instance and an EC2 with Tomcat to deliver content.
The system could look something like this:
Basically this can be the first version that you offer to customers. It has all the core features: extracting data from twitter, analysing it and display the results to the customers.
Step 6 - Automate and Measure everything
A MVP usually does not only lack features, it needs automation.
Automate as much as possible. You need to be able to concentrate on the further development and not on system operation.
Automate how to upload data to S3, stop starting the analytics by hand and write an automation script.
Start the analysis automatically and no longer by hand.
Connect the download script to the RDS database to dynamically read the list of influencers. This allows you to automatically include new influencers on customer demand.
Automate everything, create API's to ingest and store data automatically.
Then there is logging.
You need to know what you should develop next. Not only in terms of new features, it's also about fixing problems with the platform and making it faster.
Set up a system for logging and monitoring. Try to measure as much as possible.
You can log server statistics like cpu, ram, network with tools like for instance Nagios. Nagios includes a user interface for these statistics.
Log statistics for the download from twitter or upload to S3.
Log how long the Analytics process is taking and other statistics.
Log what users are doing. A simple way is to write a line in the log every time a user is using a specific function of the user interface.
Step 7 - Re-iterate
So, now your MVP is running and you automated almost everything. You have a comprehensive monitoring in place.
The system is running for some time now, you know exactly how every aspect is behaving:
You know how fast you can ingest. You know the performance of the storage and the analytics. You have a clear indication what customers are doing.
Because you implemented extensive logging, all the weaknesses in your design are visible.
What to do next?
It's time to further enhance your system. Get rid of the current weaknesses and add further functions to the system.
Getting rid of weaknesses you will optimise the overall performance and stability of the system. New features will add further value for your customers.
Implementing new features will also allow you to offer new services or products.
Some Final Words
Is it really that simple to turn an idea into a successful venture? Just by:
- Getting the data
- Selecting the right tools for analytics
- Proving your theory with science
- Figuring out your business model
- Building a minimum viable product
- Automating and measuring everything
Following these 7 steps will bring order into the chaos of building a product. They will help you to set your priorities and get the most out of your time.
However, to be honest: There is no success guarantee, despite people telling you that there is. There just isn't.
Maybe you misjudged your customers needs and you need to pivot. by changing your customer value and how you deliver. If so, these seven steps will help you find your way.
What do you think? Did I miss a crucial piece of the puzzle? Have you already done the seven steps without knowing about them?
Bio: Andreas Kretz is a Data Science & Big Data Adventurer from Germany.
Original. Reposted with permission.