A python package and analysis of the t-SNE dimensionality reduction technique.
You can install the package from pip:
pip install simple_tsne
The core functionality of the package lives in the tsne function.
The following example runs tsne on the MNIST dataset:
from simple_tsne import tsne, momentum_func
from sklearn.datasets import load_digits
import matplotlib.pyplot as plt
digits, digit_class = load_digits(return_X_y=True)
low_dim = tsne(
data=digits, # Data is mxn numpy array, each row a point
n_components=2, # Number of dim to embed to
perp=30, # Perplexity (higher for more spread out data)
n_iter=500, # Iterations to run t-SNE for
lr=100, # Learning rate
momentum_fn=momentum_func, # Function returning momentum coefficient, this one is the default update schedule
pbar=True, # Show progress bar
random_state=42 # Seed for random initialization
)
# Plot results
plt.figure()
plt.scatter(low_dim[:,0], low_dim[:,1], c=digit_class)
plt.show()