quick-tutorial.md

Introduction

In this tutorial, we present how to use pyTAGI to perform a simple regression problem. We use a 1D toy dataset and a feedforward neural network (FNN) with a simple architecture.

Define user input and data

In this simple example, we will use a 1D toy dataset. The dataset is composed of 10 training and 100 test observations and can be found in the github repository.

# User-input
num_inputs = 1
num_outputs = 1
x_train_file = "./data/toy_example/x_train_1D.csv"
y_train_file = "./data/toy_example/y_train_1D.csv"
x_test_file = "./data/toy_example/x_test_1D.csv"
y_test_file = "./data/toy_example/y_test_1D.csv"

Build Regression Model

We use a FNN with a simple architecture as defined in the RegressionMLP class (you can find the class implementation here).

# Model
net_prop = RegressionMLP()

If you want to use a different model, you can define your own class and make sure that it inherits from the NetProp class, more information in models page.

Data loader

We use the RegressionDataLoader class to load and process the data. The process_data function requires the input and output test and training files in a csv format.

# Data loader
reg_data_loader = RegressionDataLoader(num_inputs=num_inputs,
                                       num_outputs=num_outputs,
                                       batch_size=net_prop.batch_size)

data_loader = reg_data_loader.process_data(x_train_file=x_train_file,
                                           y_train_file=y_train_file,
                                           x_test_file=x_test_file,
                                           y_test_file=y_test_file)

Train and test the model

Using the regression class, we train and test the model using TAGI. Note that in order to perform the task we need to specify the number of epochs.

# Optional: Visualize the test using visualizer.py
viz = PredictionViz(task_name="regression", data_name="toy1D")

num_epochs = 50

# Train and test
reg_task = Regression(num_epochs=num_epochs,
                      data_loader=data_loader,
                      net_prop=net_prop,
                      viz=viz)
reg_task.train()
reg_task.predict(std_factor=3)

*Learn more about PredictionViz class here.

Results

Since we have created a vizualization object that we passed to the regression class, at the end of the execution we plot the results of the predictions. The black line is the true function, the red line is the predicted expected values and the red region is the confidence interval.