%pylab inline

from sklearn.manifold import TSNE

help(TSNE)

from sklearn.datasets import load_iris
from sklearn.decomposition import PCA

iris = load_iris()
X_tsne = TSNE(learning_rate=100).fit_transform(iris.data)
X_pca = PCA().fit_transform(iris.data)

figure(figsize=(10, 5))
subplot(121)
scatter(X_tsne[:, 0], X_tsne[:, 1], c=iris.target)
subplot(122)
scatter(X_pca[:, 0], X_pca[:, 1], c=iris.target)

from sklearn.datasets import fetch_20newsgroups
from sklearn.feature_extraction.text import TfidfVectorizer

categories = ['alt.atheism', 'talk.religion.misc', 'comp.graphics', 'sci.space']
newsgroups = fetch_20newsgroups(subset="train", categories=categories)
vectors = TfidfVectorizer().fit_transform(newsgroups.data)

print(repr(vectors))

from sklearn.decomposition import TruncatedSVD

X_reduced = TruncatedSVD(n_components=50, random_state=0).fit_transform(vectors)

X_embedded = TSNE(n_components=2, perplexity=40, verbose=2).fit_transform(X_reduced)

fig = figure(figsize=(10, 10))
ax = axes(frameon=False)
setp(ax, xticks=(), yticks=())
subplots_adjust(left=0.0, bottom=0.0, right=1.0, top=0.9,
                wspace=0.0, hspace=0.0)
scatter(X_embedded[:, 0], X_embedded[:, 1],
        c=newsgroups.target, marker="x")

from sklearn.datasets import fetch_mldata

# Load MNIST dataset
mnist = fetch_mldata("MNIST original")
X, y = mnist.data / 255.0, mnist.target

# Create subset and reduce to first 50 dimensions
indices = arange(X.shape[0])
random.shuffle(indices)
n_train_samples = 5000
X_pca = PCA(n_components=50).fit_transform(X)
X_train = X_pca[indices[:n_train_samples]]
y_train = y[indices[:n_train_samples]]

# Plotting function
matplotlib.rc('font', **{'family' : 'sans-serif',
                         'weight' : 'bold',
                         'size'   : 18})
matplotlib.rc('text', **{'usetex' : True})

def plot_mnist(X, y, X_embedded, name, min_dist=10.0):
    fig = figure(figsize=(10, 10))
    ax = axes(frameon=False)
    title("\\textbf{MNIST dataset} -- Two-dimensional "
          "embedding of 70,000 handwritten digits with %s" % name)
    setp(ax, xticks=(), yticks=())
    subplots_adjust(left=0.0, bottom=0.0, right=1.0, top=0.9,
                    wspace=0.0, hspace=0.0)
    scatter(X_embedded[:, 0], X_embedded[:, 1],
            c=y, marker="x")

    if min_dist is not None:
        from matplotlib import offsetbox
        shown_images = np.array([[15., 15.]])
        indices = arange(X_embedded.shape[0])
        random.shuffle(indices)
        for i in indices[:5000]:
            dist = np.sum((X_embedded[i] - shown_images) ** 2, 1)
            if np.min(dist) < min_dist:
                continue
            shown_images = np.r_[shown_images, [X_embedded[i]]]
            imagebox = offsetbox.AnnotationBbox(
                offsetbox.OffsetImage(X[i].reshape(28, 28),
                                      cmap=cm.gray_r), X_embedded[i])
            ax.add_artist(imagebox)

X_train_embedded = TSNE(n_components=2, perplexity=40, verbose=2).fit_transform(X_train)

plot_mnist(X[indices[:n_train_samples]], y_train, X_train_embedded,
           "t-SNE", min_dist=20.0)