First touch of no-code AI development with CatData

* AI introduction slide with CatData
* AI model generation and prediction workflow is covered
* No-code experiment: no math or programmings are required.
* Used in the lecture for evolutional biology course of PhD students in University of Zurich and ETH.

Transcript

1. https://catdata.ai/sign-up Last updated: Apr 22, 2020
2. 1 Very short introduction to AI/machine learning procedure n This slide explains the first steps towards AI/machine learning without math or programming • Focus on the procedure of machine learning n Use the famous machine learning dataset “Iris” • Contains data for three species of iris with four features (sepal length, sepal width, petal length and petal width) • Let’s make an AI to predict species from the length and width values • Data example • iris_training.xlsx：Iris data for training AI model • iris_test.xlsx：Measurement data to predict iris species 0: Iris setosa 1: Iris versicolor 2: Iris virginica Three species are included
3. 2 Make your account n Access via https://catdata.ai/sign-up/ n Input your information and press the ”Sign Up” button. n Authentication mail is sent to your e-mail address. Please press the link in the mail to confirm the address. https://catdata.ai/sign-in/
4. 3 Overall AI/machine learning procedure for prediction of Iris species Train model Evaluate model Prepare data to predict species Apply the trained model Predicted result Check and pre-process the data Prepare a table with species info
5. 4 Upload the data iris_training.xlsx n After signing-in, upload ”iris_training.xlsx” by clicking on the ”Create New Table” button. • https://humanome-public.s3.amazonaws.com/downloads/catdata/en/iris_training.xlsx
6. 5 Let’s start training your first AI model (1) Check whether uploaded file is correct or not (2) Save (3) Select “Training” (4) Save n Select the purpose of the uploaded table. Here we select “Training” to train a model.
7. 6 Overall AI/machine learning procedure for prediction of Iris species Train model Evaluate model Prepare data to predict species Apply the trained model Predicted result Check and pre-process the data Prepare a table with species info
8. 7 Check the distribution of measured values Check whether distributions are expected ones or not (1) Select ‘#1 iris_training” (2) Go to “Edit Action Set” n Check the data. If your data have missing values (unexpected values), they are highlighted in red. (The Iris data is well-organized data with no missing values). Just go to next.
9. 8 Pre-processing data if needed n Missing values and outliers can be removed here. To focus on a specific group (such as male/female), select the column header. Here, we need no specific selection. Go to the confirmation page. (1) Press ”Confirmation” (2) A warning appears, because “Sample ID” will be dropped as it cannot be used for model generation
10. 9 Final data check n Check the data again because this is the last time to check the distribution. • You can find that “Sample ID” is removed automatically because it has a unique id for each sample. (1) Training (2) Data is fine.
11. 10 Overall AI/machine learning procedure for prediction of Iris species Train model Evaluate model Prepare data to predict species Apply the trained model Predicted result Check and pre-process the data Prepare a table with species info
12. 11 Train a new model (1) Create New Model (2) Select a value for prediction (3) Click to start training the model Parameters to train a model. We need to tune them manually, but here we start with the recommended values.
13. 12 Wait until the training job has finished n After waiting for a few seconds or minutes (depending on your settings), the status becomes “Finished”. • Your AI model has been created! n To show how good it is, press the “Evaluation” button. (1) Press to check the model evaluation
14. 13 Overall AI/machine learning procedure for prediction of Iris species Train model Evaluate model Prepare data to predict species Apply the trained model Predicted result Check and pre-process the data Prepare a table with species info
15. 14 Evaluate the model n Given data are automatically divided into two random groups: Training and Test n The model was created on the training data and evaluated on the test data To avoid the model specializing on training data (it means loss of generality) n Confusion matrix: you can see which species are difficult to predict by the model n Accuracy: the ratio of correct samples Given data Training data (around 75%) Test data
16. 15 Which values are important for the model? n In this model (RandomForest), we can see which features are important in the the model. n From the graph, Petal length and petal width are important to discriminate between classes n On the scatter plot (bottom right), this importance is confirmed.
17. 16 Overall AI/machine learning procedure for prediction of Iris species Train model Evaluate model Prepare data to predict species Apply the trained model Predicted result Check and pre-process the data Prepare a table with species info
18. 17 Upload the test data iris_test.xlsx n Upload ” iris_test.xlsx” after clicking the ”New Table” button. https://humanome-public.s3.amazonaws.com/downloads/catdata/en/iris_test.xlsx
19. 18 Set the purpose of the uploaded table to “Prediction” (1) Check whether uploaded file is correct or not (2) Save (3) Select “Prediction” (4) Save n Select the purpose of the uploaded table. Here we select “Prediction” to evaluate a model.
20. 19 Check whether the selected table is correct or not Check whether distributions are expected or not (1) Select ‘#2 iris_test” (2) Go to “Prediction” n Check the data. This is the same as ”Training”. If it’s ok, just click ”Prediction”
21. 20 Select a model to be applied to the uploaded table (1) Select ‘#1 iris_training” n Check the data. This is the same as ”Training”. If it’s ok, just click ”Prediction” (2) Press the “Start” button List of models created from #1 table
22. 21 The prediction is done. Go to the detail of the predicted results n Status becomes ”Finished” after a few seconds. n Then, press the ”Result” button to check the details (next page)
23. 22 Predicted results are added to the test data n The green highlighted column is the predicted result. n Comparing with Sample ID, you can see if the results are correct or not S: Iris setosa, VE: Iris versicolor, VI: Iris virginica
24. 23 Summary n This is a very short demonstration of AI/machine learning. n I focused on the workflow of AI/ML model generation and testing. I skipped some important words and measures. n Try to change models and parameters and see how the results differ. n If you have your own experimental data, try to make your model from the data. n Functions on the website are limited. If you would like to do a more detailed analysis, you should learn R or Python.