I am teaching the AI and Machine Learning Unit for the first time. Everything had been going beautifully until we got to Lesson 15 in which students choose a dataset, create a model card, train the A, and then import the model into app studio. None of us were able to create a meaningful app. The data sets just didn’t seem to have enough or the right kind of information. For instance, one students wanted to make an app to recommend a movie. They couldn’t create a model to do that. Does anyone have any examples of models that work well from the given data sets? Is it possible to import your own dataset? Also, are the students supposed to fill out a model card as they had done previously?
Hi @beniec ,
Welcome to the forum! It sounds like you are working on the mini-project in Unit 7 of Discoveries titled “Make a Machine Learning App”. I took a look at the lesson plan and it does state that step 1 is that “Students will choose a dataset from the provided list in AI Lab”. I don’t see a movie related list provided so a movie recommender is not possible with the provided datasets (6 are provided). Yes, there is an option to upload a csv or dataset but you would want to make sure the dataset has all the required information for prediction. That is why datasets that would work are provided. Step 2 in the lesson plan is to train their model. When they do this and see the results of their training, they can be sure they are choosing options that will lead to good predictions which may help in making meaningful apps. And step 3 is to make a model card. So yes, students make a model card as they have done previously. The lesson activities walk you through each of these steps.
Perhaps you can reply with more details such as when a student chose the Billionaire dataset and which columns they chose for “Predict” and “based on” and others can reply with suggestions to improve results?
~Michelle
Thank you for your response!
I am working on the standalone machine learning unit. In Lesson 15 there are 18 data sets, so it must be slightly different that Discoveries.
The kids and I definitely knew how to train the model, but we struggled with coming up with an app that used that data to provide any useful information. Let’s say I have an app that can recommend a hiking trail based on time of year, trail length, and trail difficulty. That’s the kind of thing we had in mind, but none of use could figure out a way to do this with the provided datasets. For instance, what on earth can you do with the millionaire data?
I feel like I am just missing something with this lesson and would love to see some examples of apps using those datasets.
Edit.
MODEL CARD
In the lesson plan, Step 3 reads:
Step 3 - Save Your Model, Create a Model Card: Students will create a model card in AI Lab and save their model. They will also record some of their answers on their project guide.
In the activity guide, step 3 reads:
Step 3: Save your Model, Create a Model Card
Save your model in AI Lab. Before continuing to App Lab, answer the following questions from your Model Card:
I didn’t see a printed model card template but I also didn’t see a place in AI Lab to create a model card.
LABLE & FEATURES CHOSEN
Here is the code from one student using the millionaire dataset:
var data = {};
onEvent(“startbutton”, “click”, function( ) {
setScreen(“screen1”);
});
onEvent(“Billionaire_predict”, “click”, function() {
addPair(data, “WealthInherited”, getText(“WealthInherited_dropdown”));
addPair(data, “WealthIndustry”, getText(“WealthIndustry_dropdown”));
addPair(data, “NetWorthInBillions”, getText(“NetWorthInBillions_input”));
setText(“Billionaire_prediction”, ‘’);
getPrediction(“Billionaire”, “YitNH7Tr69oD”, data, function(value) {
setText(“Billionaire_prediction”, value);
});
});
Thanks for the clarification @beniec . So, here are my thoughts. The type of app you described, “recommend a hiking trail based on time of year, trail length, and trail difficulty” really isn’t a predictive app. That is the type of app my students make in AP CSP by filtering data. You would take a dataset about hiking trails and the user would filter for season, length, and difficulty and show a list of results. The apps you make in the AI Machine Learning tool are predictive. They are more of a “determine the likelihood based on this info” or “guess the answer based on this info”. I quickly made a salary predictor app based on some different criteria. I did add a little code to make the output text a little more descriptive.
A tip to using the datasets and determining what kind of app you can make is to read the “Potential Uses” when you first click on the dataset. Remember, you can add code. So for example in the Abalone dataset, the user could guess a weight and the code you program could grab their guess and run it through an if statement to output whether they guessed close or not (similar to how my sample app).
And yes, you can use Kaggle or other datasets. I used this dataset to make a GPA predictor. Our school blocks Kaggle so what I have done is download a few datasets and then put them in a shared Google Drive for kids to access.
Hope that helps but also hope others share their apps.
~Michelle
1 Like
Thank you! This is very helpful. In my past life. I was a backend website developer, so I think I as looking at this incorrectly. It’s making more sense to me now. Thank you!
1 Like