Tips

Filters

• Hover over the categories and subcategories the Filters section to view tooltips or view them on the Methodology page.
• Select a category in the Filters section to view its subcategories, then select a subcategory to update the visualization.
• There are breadcrumbs immediately above the visualization that indicate the category and subcategory you have currently selected. They will update if you update your filters or click through to another subcategory from the search results.

Flowchart

• You can only view one flowchart at a time. (Some of them are quite complex, so this was a design decision.)
• This tool is highly interactive. If you experience bugs, it may be because your extensions are interfering. Copy the app url and paste it into an incognito browser. That's what I frequently need to do with my apps because of how many extensions I'm saddling my browser with in my work instance.
• You can click on the decision nodes (orange diamonds) or model nodes (blue rectangles) to open modals with more information. The decision node tooltips take an 'explain it to me like I'm 5 (ELIF)' approach, to make the tool as accessible as possible for those with less experience. You can usually just click/tap away to close them, but each modal has a close button for when the browser gets finicky.
• The model cards have a lot of extra information packed into them, including a description of the model, example use cases, difficulty level, and a link to a Python and R tutorial. If I couldn't find an R tutorial, I noted that you can use the reticulate library to use Python libraries in R. And sometimes an R library has an option to use a model that the tutorial doesn't hit on. In those cases, I included a link to the tutorial. You can always crack open the documentation to see, for example, how to modify a tutorial about the SPADE model to use PrefixSpan instead. Click the x or anywhere outside the modal to close it. And I cover my methodology in assigning a difficulty level on the Methodology page.
• If a search match is in the label of a decision node (diamond) or model node (box), the app will highlight the node but not the keyword. So if you fire up a modal and nothing is highlighted, the keyword is probably in the label. Otherwise, you found a 🦟 and you could do me a solid by letting me know! 👊
• You can drag tooltips to reposition them in your browser. Just click/tap anywhere in the header to select the handle.
• If you want to verify a flowchart strategy, you can grab a screenshot of it, drop it into an AI tool with a reasoning model (which most offer at this point), and ask it to give you specific recommendations for your challenge. These models can help you differentiate between multiple options and provide you with more constraints to consider than a tool like this could capture.
• Some of the model names were long (e.g., 'Convolutional Neural Networks (CNN) with RNN Decoder') so I used shortened names for those labels. I include the full model name in the modal. For most modals this means the model name will be duplicated. Sometimes sacrifices must be made. 🤌
• You can zoom using your mouse wheel (or its equivalent) and pan either the canvas by dragging it. You can also move individual nodes, but you shouldn't have to. However, zooming and panning can cause modals to be cut off. If that happens, just zoom out a bit or drag the canvas so that the node isn't so close to the edge and try again.
• You can adjust the height of the canvas by dragging the handle in the bottom border...for you divas with larger-than-life screens. 🙆‍♀️🙆‍♂️
• If a tutorial is too advanced for you, AI is a wonderful tutor. My fave tool is Claude—which at the time of writing (Feb 2025) can't follow links—so I just paste the tutorial into the chat window and ask it to first explain it to me at a high level. This helps me understand the overall strategy of a particular model. (You can link to it if you're using ChatGPT, but sometimes it hallucinates content in linked content so user beware.) If the concepts are heady or the explanation is too technical, I will explicitly ask it to use everyday analogies or explain it like I'm five. Then I ask it to walk me through how to perform the analysis step by step. You can also provide it the first few rows of your data—or contrived data with the same column labels as your dataset uses. At the time of writing, Claude is much better at this than ChatGPT. I usually need to keep asking ChatGPT to give me steps one at a time if I start running into problems. One time it still gave me 7 steps but called them Step 1a, Step 1b, etc. You can't make this stuff up. 🤦‍♀️
• Although the best experience will be on desktop, it is fully responsive for mobile.
• There may be an 🐰🥚 in the app. Let me know if you find it! 👀

Search

• When using the search field, if there are multiple search results, you will see nested results in the search panel. Clicking on a category will reveal the subcategory (or subcategories) with a match. Selecting a subcategory will update the flowchart for that subcategory and update the breadcrumbs.
• In the chart, all nodes with a match will be highlighted and the nodes that don't contain a match will drop in opacity. When you open a tooltip or model card, the search term will be highlighted in orange to make it more easily scannable.
• Keyword matches won't be highlighted in node labels, just the node itself. So if you open a decision node and don't see anything highlighted, check its label. This is only relevant to decision nodes because the model name in the label is also in the card.
• If you want to keep your search active but remove the search panel from your screen, just select the minimize button in the upper-right corner. You can go back into the search by selecting the magnifying glass icon.
• Clearing the search field will return the nodes to their default formatting.
• Sometimes the decision node tooltips get a little buggy on small devices in portrait mode (i.e., vertical) after performing a search. If you flip your phone to landscape and then back to portrait, that should resolve the issue. Javascript can be such a 🎭👸.