Iris Classification Model using Logistic Regression and Artificial Neural Network
In this project, I used logistic regression and artificial neural network to classify the iris dataset on 4 features: sepal length, sepal width, petal length, and petal width. The iris dataset was loaded using scikit-learn and the model was trained using PyTorch.
First, I loaded the iris dataset using scikit-learn’s load_iris function. Then, I split the data into training and test sets using train_test_split function in the ratio of 80:20. After that, I trained a logistic regression model and an artificial neural network model on the training data. Finally, I evaluated the performance of both models on the test data by storing the loss and accuracy history and then plotting them later on.
To prepare the data for training, I first split it into training and test sets using train_test_split function from scikit-learn. Then, I standardized the data using StandardScaler and converted the data into PyTorch tensors. After that, I created data loaders for training and test data using DataLoader class from PyTorch.
I also defined two models: a logistic regression model and a neural network model. The logistic regression model has one linear layer with 4 input features and 3 output classes. The neural network model has two fully connected layers with a ReLU activation function in between(I did look out for other activation functions but ReLU worked the best imo).
To train the models, I defined a train function that takes in a model, a loss function (criterion), an optimizer, and the number of epochs as inputs(2000 epochs worked best for me). Inside the function, I looped over the number of epochs and for each epoch, I looped over the batches of data in the training data loader. For each batch, I performed a forward pass to compute the predicted outputs (y_pred) and the loss. Then, I performed a backward pass to compute the gradients and update the model parameters using the optimizer.
After each epoch, I computed the loss and accuracy on the test data and appended them to loss_history and accuracy_history lists respectively.
I trained both the logistic regression model and the neural network model using this train function with CrossEntropyLoss as the loss function and SGD as the optimizer.
After training was complete, I plotted the loss and accuracy curves for both models using matplotlib. The loss curve showed how the loss changed over the epochs during training, while the accuracy curve showed how the accuracy on the test data changed over the epochs.
For data split 80:20 and batch size = 8 and random state = 1 and over 2000 epochs
For data split 80:20 and batch size = 16 and random state = 0 and over 1000 epochs
