Announcing the Annielytics Machine Learning Model Picker App

When I first set out to build free apps to help the AI community, the trifecta I had in mind included:

• An interactive AI timeline to help those in the field keep pace with the dizzying rate of innovation, the AI arms race, the legal ramifications of some of this innovation, and AI fails (to name a few). You can access that here and learn how to use it here.
• An interactive AI Strategy app that guides practitioners through the harrowing process of comparing the performance of AI models. Model providers make a lot of performance claims and, frankly, it’s rare for any of them to be substantiated via these leaderboards, a growing issue I address in this post. You can access that app here and learn how to use it here.
• An interactive Machine Learning Model Picker that guides data scientists and stakeholders through the even more harrowing process of picking the best model for the task at hand. Now you can access that here. 🙌

TL;DR

You can use my Machine Learning Model Picker completely free by choosing a category, then a subcategory, then following the flowchart until you land on a model. If you’re not clear on a question, select the diamond decision node to get further explanation in simple terms (hence the ‘ELI5‘—or ‘explain it like I’m 5‘ heading). Finally, learn more about the model by selecting the model to open a model card with a cornucopia of information and links.

How To Use It

Step 1: Pick a Category

There are six categories of models to choose from:

1. Association
2. Generation
3. Order
4. Pattern Discovery
5. Prediction
6. Preprocessing

You can hover over a category to learn more about it and get example tasks—or fire up the app’s Methodology page and scroll down to the Category Definitions section, if you’re on a touch device.

Step 2: Pick a Subcategory

Each category has anywhere from two to six subcategories. And each has a tooltip just like its parent category. Since I made this app responsive for mobile devices—and updated my other apps to be responsive as well—I couldn’t get these tooltips to behave across all devices, operating systems, and browsers, so I added handles to them in the upper-left corner so that you can drag them where you want them.

Truth be told, you can drag them by grabbing anywhere in the blue header. The icon just serves as a recognizable visual cue.

I also programmed them to close by pressing the Escape key or clicking outside the tooltip. The Escape key is reliable; clicking outside the tooltip…not so much. So I added a close button as well.

Aside: I’ve gained a new appreciation for front-end developers in building these apps. I found getting different JavaScript frameworks to play nicely together and cooperate with my CSS infinitely more difficult than getting Python libraries to work in harmony.

Step 3: Generate Flowchart

The Classification subcategory is one of the more complex flowcharts because there are quite a few models to choose from, so don’t let it scare you.

Decision Node Tooltips

Each decision node (the orange diamonds) has a question with a ‘yes’ or ‘no’ response. You can click/tap on a node to learn more. I added these because, if you don’t understand the question, you run the risk of choosing a suboptimal model for your project. They also have handles.

Model Cards

When you wend your way down to a model, you can click/tap it to learn more. This is really where the magic happens. The model card has six sections:

• Model Name: I added this for when the name was too long for the title section.
• Description: This is a brief description of the model.
• Use Cases: I include example use cases for each model. To get more, I highly recommend asking your fave AI tool. There are many. I tried to vary them as much as possible as there is a total of 102 model cards, and I didn’t want these use cases to get too repetitive. I relied heavily on Claude for these use cases but then used ChatGPT’s reasoning model to double check Claude’s work (and sometimes vice versa).
• Alternate Models: I added this section later because I didn’t want to give the impression that the models included in this tool are the only models you can use. There are many to choose from, and sometimes differentiating them was a challenge.
• Difficulty Level: These levels are based on six factors, which I documented on my Methodology page. I used Claude to generate the scores.
• Tutorials: Machine Learning is dominated by Python libraries. However, R holds its own, and its user base is passionate so I included it as well. In cases where an R package didn’t exist and/or didn’t have a good tutorial, the reticulate package allows users to to run Python libraries in RStudio. I so wish Python had a dplyr library. It’s so much cleaner than Python’s Pandas. (I said what I said.)

You can get a feel for what scanning the flowchart is like in this short demo video.

Other Features

Search

The tool has a search feature that extends beyond just the flowchart you’re currently on. Let’s say I want to search for ‘image’ to see all the instances where machine learning tasks are image leaning.

There’s quite a bit to unpack in this screenshot, so I’ll summarize the features as succinctly as possible.

• You will see a list of categories that contain matches. Selecting one will reveal the subcategory/ies with matches. Clicking on one of them will open that subcat’s flowchart.
• All decision and model nodes that contain a match will be highlighted for ease of scanning.
• The search term will also be highlighted in the decision tooltip or model card. (I do this for all my apps because I know how much people hate to read. 😉)
• If the search panel is in your way, you can minimize it by selecting the button in the upper-right corner. You can go back in any time by selecting the magnifying glass icon. It will be accessible to you, with the highlighted nodes, until you close out of the search.

Chart Resizer Handle

The chart canvas has a resizer bar along the bottom. Click or tap and drag to adjust the height to better accommodate your screen size.

Zoom

You can use your mouse wheel or pinch and drag on touch devices to zoom in and out. This is particularly helpful after resizing the chart. I slowed the zoom feature down for the mouse wheel to minimize erratic repositioning as you can experience with the default setting. I also allotted more padding outsize the vis so you can scroll without activating the zoom unintentionally.

Auxiliary Resources

In the footer are links to a tips page with so. many. tips. I also have a methodology, resources, and feedback page. Please let me know if you run into any bugs, broken links, or inaccurate/outdated information.

Updates to My Other Apps

When I created this app, I lightened up the design a bit, after figuring out how to replicate my WordPress header in the app and the supplementary pages that sit outside WordPress. As I mentioned earlier, I also made this app responsive for mobile users. That was quite the learning curve, but once I got the hang of it, I was creating custom breakpoints to customize the design as you even flipped your phone from portrait to landscape.

Afterwards I decided to apply the same changes to my other apps. The AI Strategy app also has a minimize feature now for the orientation slider as it took up quite a bit of screen real estate on mobile.

And now these apps have given me years’ worth of content.

Easter Egg

I had some fun adding one to the app. Let me know if you find it. 👀

Image credit: Bruce Mars

Music credit: Dvir Silver