Customer Segmentation Pipeline Prototype

optics.py

"""
Copyright 2015 Ronald J. Nowling

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
"""

import sys
import numpy as np
import matplotlib.pyplot as plt
import csv

from sklearn.decomposition import PCA
from sklearn.cluster import k_means
from scipy.stats import rankdata
from sklearn.feature_extraction.text import TfidfTransformer
from sklearn.metrics import confusion_matrix
from sklearn.naive_bayes import MultinomialNB

def read_matrix(flname):
    ratings = []
    max_user_id = 0
    max_item_id = 0
    
    with open(flname) as fl:
        reader = csv.reader(fl, delimiter="\t")
        for rec in reader:
            user_id = int(rec[0])
            item_id = int(rec[1])
            rating = float(rec[2])
            ratings.append((user_id, item_id, rating))
            max_user_id = max(max_user_id, user_id)
            max_item_id = max(max_item_id, item_id)

    features = np.zeros((max_user_id, max_item_id))
    for user_id, item_id, rating in ratings:
        features[user_id - 1, item_id - 1] = rating


    return features

def read_items(flname):
    items = []

    with open(flname) as fl:
        reader = csv.reader(fl, delimiter="|")
        for rec in reader:
            items.append(rec[1])

    return items

def plot_explained_var_ratios(dirname, pca):
    plt.plot(pca.explained_variance_ratio_[:20], "b.-")
    plt.xlabel("Component", fontsize=16)
    plt.ylabel("Explained Variance Ratio", fontsize=16)
    plt.ylim([0.0, 1.0])
    plt.savefig(dirname + "/pca_explained_var_ratio.pdf", DPI=200)

def plot_clusters(dirname, proj_features, features):
    plt.clf()
    plt.hold(True)
    ks = [5, 10, 25, 50, 100, 125, 150, 200]
    proj_costs = []
    full_costs = []
    for k in ks:
        centroids, labels, inertia = k_means(proj_features, k)
        proj_costs.append(inertia)
        
        centroids, labels, inertia = k_means(features, k)
        full_costs.append(inertia)

    plt.plot(ks, np.array(proj_costs) / max(proj_costs), "b.-", label="Projected")
    plt.plot(ks, np.array(full_costs) / max(full_costs), "r.-", label="Full")
    plt.legend(loc="upper right")
    plt.xlabel("Centers", fontsize=16)
    plt.ylabel("Inertia", fontsize=16)
    plt.savefig(dirname + "/kmeans_inertia.pdf", DPI=200)

def plot_confusion_matrix(dirname, cm, title='Confusion matrix', cmap=plt.cm.Blues):
    plt.clf()
    plt.imshow(cm, interpolation='nearest')
    plt.title(title)
    plt.colorbar()
    plt.tight_layout()
    plt.ylabel('True label')
    plt.xlabel('Predicted label')
    plt.savefig(dirname + "/confusion_matrix.pdf", DPI=200)
            
if __name__ == "__main__":
    movielens_dir = sys.argv[1]
    
    data_fl = movielens_dir + "/u.data"
    items_fl = movielens_dir + "/u.item"

    outdir = "figures"
    
    features = read_matrix(data_fl)
    print features.shape

    sparsity = float(np.count_nonzero(features)) / (features.shape[0]
                                                    * features.shape[1])

    print "Sparsity", sparsity

    pca = PCA()
    projected = pca.fit_transform(features)

    plot_explained_var_ratios(outdir, pca)

    n_pcs = 10

    plot_clusters(outdir, projected[:, :n_pcs], features)

    _, labels, _ = k_means(projected[:, :n_pcs], 25)

    nb = MultinomialNB()
    nb.fit(features, labels)
    predicted = nb.predict(features)
    cm = confusion_matrix(labels, predicted)
    plot_confusion_matrix(outdir, cm)
    
    scaler = TfidfTransformer()
    scaled = scaler.fit_transform(features, labels).todense()
    print scaled.shape
    
    items = read_items(items_fl)

    for i in xrange(25):
       members = scaled[labels == i, :]
       print i, members.shape
       avg_rank = np.mean(members, axis=0)
       ranks = rankdata(avg_rank)

       rank_idx = np.argsort(-1.0 * ranks)
       for j in rank_idx[:10]:
           print "\t", items[j], avg_rank[0, j]