Member-only story
Use sklearn’s neural network to predict on podcast listening times
It doesn’t matter what I am engaged in, I always try to fit Kaggle’s monthly playground competition into my schedule. This month required the data scientist to predict the length of time that an individual would listen to a podcast.
While I have used a linear regression model to make predictions in the past I opted to try out a nonlinear model in this competition.
MLPRegressor in scikit-learn is a multi-layer perceptron (MLP) regressor, which is a type of neural network used for regression tasks. It learns complex relationships between input features and a continuous target variable.
Key Features of MLPRegressor
- Hidden Layers: You can specify the number of hidden layers and neurons in each layer using hidden_layer_sizes=(100,), where 100 represents the number of neurons in the single hidden layer.
- Activation Functions: Supports activation functions like ReLU, tanh, logistic, and identity.
- Optimization Solvers: Uses Adam, SGD, or LBFGS for weight optimization.
- Regularization: Includes L2 regularization (alpha parameter) to prevent overfitting.
- Learning Rate: Can be constant, adaptive, or inverse scaling.
Sklearn’s MLPRegressor can be used when:-
- The data has non-linear relationships that traditional regression models struggle with.
- When you need a flexible model that can learn complex patterns.
- When you have enough data to train a neural network effectively.
I haven’t used sklearn’s neural network for a while, so I thought it would be a good refresher for me to employ this model. I write the code in Python, using Kaggle’s Jupyter Notebook, and stored it in my account with Kaggle.
The first thing that I did after creating the Jupyter Notebook was to import the libraries that I would need to execute the program, being:-
- Numpy to perform numeric computations on the data,
- Pandas to perform data processing on the data,
- Os to go into the computer’s operating system,
