Show Sidebar Hide Sidebar

Out-of-core classification of text documents in Scikit-learn

This is an example showing how scikit-learn can be used for classification using an out-of-core approach: learning from data that doesn’t fit into main memory. We make use of an online classifier, i.e., one that supports the partial_fit method, that will be fed with batches of examples. To guarantee that the features space remains the same over time we leverage a HashingVectorizer that will project each example into the same feature space. This is especially useful in the case of text classification where new features (words) may appear in each batch.

The dataset used in this example is Reuters-21578 as provided by the UCI ML repository. It will be automatically downloaded and uncompressed on first run.

The plot represents the learning curve of the classifier: the evolution of classification accuracy over the course of the mini-batches. Accuracy is measured on the first 1000 samples, held out as a validation set. To limit the memory consumption, we queue examples up to a fixed amount before feeding them to the learner.

New to Plotly?

Plotly's Python library is free and open source! Get started by downloading the client and reading the primer.
You can set up Plotly to work in online or offline mode, or in jupyter notebooks.
We also have a quick-reference cheatsheet (new!) to help you get started!

Version

In [1]:
import sklearn
sklearn.__version__
Out[1]:
'0.18'

Imports

In [2]:
import plotly.graph_objs as go
import plotly.plotly as py
from plotly import tools

from __future__ import print_function

from glob import glob
import itertools
import os.path
import re
import tarfile
import time

import numpy as np
from matplotlib import rcParams

from sklearn.externals.six.moves import html_parser
from sklearn.externals.six.moves import urllib
from sklearn.datasets import get_data_home
from sklearn.feature_extraction.text import HashingVectorizer
from sklearn.linear_model import SGDClassifier
from sklearn.linear_model import PassiveAggressiveClassifier
from sklearn.linear_model import Perceptron
from sklearn.naive_bayes import MultinomialNB
In [3]:
def _not_in_sphinx():
    # Hack to detect whether we are running by the sphinx builder
    return '__file__' in globals()
In [4]:
class ReutersParser(html_parser.HTMLParser):
    """Utility class to parse a SGML file and yield documents one at a time."""

    def __init__(self, encoding='latin-1'):
        html_parser.HTMLParser.__init__(self)
        self._reset()
        self.encoding = encoding

    def handle_starttag(self, tag, attrs):
        method = 'start_' + tag
        getattr(self, method, lambda x: None)(attrs)

    def handle_endtag(self, tag):
        method = 'end_' + tag
        getattr(self, method, lambda: None)()

    def _reset(self):
        self.in_title = 0
        self.in_body = 0
        self.in_topics = 0
        self.in_topic_d = 0
        self.title = ""
        self.body = ""
        self.topics = []
        self.topic_d = ""

    def parse(self, fd):
        self.docs = []
        for chunk in fd:
            self.feed(chunk.decode(self.encoding))
            for doc in self.docs:
                yield doc
            self.docs = []
        self.close()

    def handle_data(self, data):
        if self.in_body:
            self.body += data
        elif self.in_title:
            self.title += data
        elif self.in_topic_d:
            self.topic_d += data

    def start_reuters(self, attributes):
        pass

    def end_reuters(self):
        self.body = re.sub(r'\s+', r' ', self.body)
        self.docs.append({'title': self.title,
                          'body': self.body,
                          'topics': self.topics})
        self._reset()

    def start_title(self, attributes):
        self.in_title = 1

    def end_title(self):
        self.in_title = 0

    def start_body(self, attributes):
        self.in_body = 1

    def end_body(self):
        self.in_body = 0

    def start_topics(self, attributes):
        self.in_topics = 1

    def end_topics(self):
        self.in_topics = 0

    def start_d(self, attributes):
        self.in_topic_d = 1

    def end_d(self):
        self.in_topic_d = 0
        self.topics.append(self.topic_d)
        self.topic_d = ""


def stream_reuters_documents(data_path=None):
    """Iterate over documents of the Reuters dataset.

    The Reuters archive will automatically be downloaded and uncompressed if
    the `data_path` directory does not exist.

    Documents are represented as dictionaries with 'body' (str),
    'title' (str), 'topics' (list(str)) keys.

    """

    DOWNLOAD_URL = ('http://archive.ics.uci.edu/ml/machine-learning-databases/'
                    'reuters21578-mld/reuters21578.tar.gz')
    ARCHIVE_FILENAME = 'reuters21578.tar.gz'

    if data_path is None:
        data_path = os.path.join(get_data_home(), "reuters")
    if not os.path.exists(data_path):
        """Download the dataset."""
        print("downloading dataset (once and for all) into %s" %
              data_path)
        os.mkdir(data_path)

        def progress(blocknum, bs, size):
            total_sz_mb = '%.2f MB' % (size / 1e6)
            current_sz_mb = '%.2f MB' % ((blocknum * bs) / 1e6)
            if _not_in_sphinx():
                print('\rdownloaded %s / %s' % (current_sz_mb, total_sz_mb),
                      end='')

        archive_path = os.path.join(data_path, ARCHIVE_FILENAME)
        urllib.request.urlretrieve(DOWNLOAD_URL, filename=archive_path,
                                   reporthook=progress)
        if _not_in_sphinx():
            print('\r', end='')
        print("untarring Reuters dataset...")
        tarfile.open(archive_path, 'r:gz').extractall(data_path)
        print("done.")

    parser = ReutersParser()
    for filename in glob(os.path.join(data_path, "*.sgm")):
        for doc in parser.parse(open(filename, 'rb')):
            yield doc

Create the vectorizer and limit the number of features to a reasonable maximum

In [5]:
vectorizer = HashingVectorizer(decode_error='ignore', n_features=2 ** 18,
                               non_negative=True)


# Iterator over parsed Reuters SGML files.
data_stream = stream_reuters_documents()

# We learn a binary classification between the "acq" class and all the others.
# "acq" was chosen as it is more or less evenly distributed in the Reuters
# files. For other datasets, one should take care of creating a test set with
# a realistic portion of positive instances.
all_classes = np.array([0, 1])
positive_class = 'acq'

# Here are some classifiers that support the `partial_fit` method
partial_fit_classifiers = {
    'SGD': SGDClassifier(),
    'Perceptron': Perceptron(),
    'NB Multinomial': MultinomialNB(alpha=0.01),
    'Passive-Aggressive': PassiveAggressiveClassifier(),
}


def get_minibatch(doc_iter, size, pos_class=positive_class):
    """Extract a minibatch of examples, return a tuple X_text, y.

    Note: size is before excluding invalid docs with no topics assigned.

    """
    data = [(u'{title}<br><br>{body}'.format(**doc), pos_class in doc['topics'])
            for doc in itertools.islice(doc_iter, size)
            if doc['topics']]
    if not len(data):
        return np.asarray([], dtype=int), np.asarray([], dtype=int)
    X_text, y = zip(*data)
    return X_text, np.asarray(y, dtype=int)


def iter_minibatches(doc_iter, minibatch_size):
    """Generator of minibatches."""
    X_text, y = get_minibatch(doc_iter, minibatch_size)
    while len(X_text):
        yield X_text, y
        X_text, y = get_minibatch(doc_iter, minibatch_size)


# test data statistics
test_stats = {'n_test': 0, 'n_test_pos': 0}

# First we hold out a number of examples to estimate accuracy
n_test_documents = 1000
tick = time.time()
X_test_text, y_test = get_minibatch(data_stream, 1000)
parsing_time = time.time() - tick
tick = time.time()
X_test = vectorizer.transform(X_test_text)
vectorizing_time = time.time() - tick
test_stats['n_test'] += len(y_test)
test_stats['n_test_pos'] += sum(y_test)
print("Test set is %d documents (%d positive)" % (len(y_test), sum(y_test)))


def progress(cls_name, stats):
    """Report progress information, return a string."""
    duration = time.time() - stats['t0']
    s = "%20s classifier : \t" % cls_name
    s += "%(n_train)6d train docs (%(n_train_pos)6d positive) " % stats
    s += "%(n_test)6d test docs (%(n_test_pos)6d positive) " % test_stats
    s += "accuracy: %(accuracy).3f " % stats
    s += "in %.2fs (%5d docs/s)" % (duration, stats['n_train'] / duration)
    return s


cls_stats = {}

for cls_name in partial_fit_classifiers:
    stats = {'n_train': 0, 'n_train_pos': 0,
             'accuracy': 0.0, 'accuracy_history': [(0, 0)], 't0': time.time(),
             'runtime_history': [(0, 0)], 'total_fit_time': 0.0}
    cls_stats[cls_name] = stats

get_minibatch(data_stream, n_test_documents)
# Discard test set

# We will feed the classifier with mini-batches of 1000 documents; this means
# we have at most 1000 docs in memory at any time.  The smaller the document
# batch, the bigger the relative overhead of the partial fit methods.
minibatch_size = 1000

# Create the data_stream that parses Reuters SGML files and iterates on
# documents as a stream.
minibatch_iterators = iter_minibatches(data_stream, minibatch_size)
total_vect_time = 0.0

# Main loop : iterate on mini-batches of examples
for i, (X_train_text, y_train) in enumerate(minibatch_iterators):

    tick = time.time()
    X_train = vectorizer.transform(X_train_text)
    total_vect_time += time.time() - tick

    for cls_name, cls in partial_fit_classifiers.items():
        tick = time.time()
        # update estimator with examples in the current mini-batch
        cls.partial_fit(X_train, y_train, classes=all_classes)

        # accumulate test accuracy stats
        cls_stats[cls_name]['total_fit_time'] += time.time() - tick
        cls_stats[cls_name]['n_train'] += X_train.shape[0]
        cls_stats[cls_name]['n_train_pos'] += sum(y_train)
        tick = time.time()
        cls_stats[cls_name]['accuracy'] = cls.score(X_test, y_test)
        cls_stats[cls_name]['prediction_time'] = time.time() - tick
        acc_history = (cls_stats[cls_name]['accuracy'],
                       cls_stats[cls_name]['n_train'])
        cls_stats[cls_name]['accuracy_history'].append(acc_history)
        run_history = (cls_stats[cls_name]['accuracy'],
                       total_vect_time + cls_stats[cls_name]['total_fit_time'])
        cls_stats[cls_name]['runtime_history'].append(run_history)

        if i % 3 == 0:
            print(progress(cls_name, cls_stats[cls_name]))
    if i % 3 == 0:
        print('\n')
Test set is 469 documents (49 positive)
  Passive-Aggressive classifier : 	   969 train docs (   155 positive)    469 test docs (    49 positive) accuracy: 0.940 in 1.38s (  700 docs/s)
          Perceptron classifier : 	   969 train docs (   155 positive)    469 test docs (    49 positive) accuracy: 0.923 in 1.39s (  697 docs/s)
                 SGD classifier : 	   969 train docs (   155 positive)    469 test docs (    49 positive) accuracy: 0.936 in 1.39s (  695 docs/s)
      NB Multinomial classifier : 	   969 train docs (   155 positive)    469 test docs (    49 positive) accuracy: 0.896 in 1.42s (  684 docs/s)


  Passive-Aggressive classifier : 	  3731 train docs (   466 positive)    469 test docs (    49 positive) accuracy: 0.966 in 3.67s ( 1016 docs/s)
          Perceptron classifier : 	  3731 train docs (   466 positive)    469 test docs (    49 positive) accuracy: 0.934 in 3.67s ( 1015 docs/s)
                 SGD classifier : 	  3731 train docs (   466 positive)    469 test docs (    49 positive) accuracy: 0.934 in 3.68s ( 1014 docs/s)
      NB Multinomial classifier : 	  3731 train docs (   466 positive)    469 test docs (    49 positive) accuracy: 0.902 in 3.70s ( 1008 docs/s)


  Passive-Aggressive classifier : 	  6593 train docs (   811 positive)    469 test docs (    49 positive) accuracy: 0.972 in 5.97s ( 1104 docs/s)
          Perceptron classifier : 	  6593 train docs (   811 positive)    469 test docs (    49 positive) accuracy: 0.945 in 5.97s ( 1103 docs/s)
                 SGD classifier : 	  6593 train docs (   811 positive)    469 test docs (    49 positive) accuracy: 0.962 in 5.98s ( 1103 docs/s)
      NB Multinomial classifier : 	  6593 train docs (   811 positive)    469 test docs (    49 positive) accuracy: 0.913 in 6.00s ( 1099 docs/s)


  Passive-Aggressive classifier : 	  9531 train docs (  1163 positive)    469 test docs (    49 positive) accuracy: 0.962 in 8.26s ( 1154 docs/s)
          Perceptron classifier : 	  9531 train docs (  1163 positive)    469 test docs (    49 positive) accuracy: 0.951 in 8.26s ( 1153 docs/s)
                 SGD classifier : 	  9531 train docs (  1163 positive)    469 test docs (    49 positive) accuracy: 0.959 in 8.26s ( 1153 docs/s)
      NB Multinomial classifier : 	  9531 train docs (  1163 positive)    469 test docs (    49 positive) accuracy: 0.923 in 8.28s ( 1150 docs/s)


  Passive-Aggressive classifier : 	 12353 train docs (  1591 positive)    469 test docs (    49 positive) accuracy: 0.953 in 10.50s ( 1176 docs/s)
          Perceptron classifier : 	 12353 train docs (  1591 positive)    469 test docs (    49 positive) accuracy: 0.951 in 10.51s ( 1175 docs/s)
                 SGD classifier : 	 12353 train docs (  1591 positive)    469 test docs (    49 positive) accuracy: 0.962 in 10.51s ( 1175 docs/s)
      NB Multinomial classifier : 	 12353 train docs (  1591 positive)    469 test docs (    49 positive) accuracy: 0.934 in 10.53s ( 1173 docs/s)


  Passive-Aggressive classifier : 	 15296 train docs (  1916 positive)    469 test docs (    49 positive) accuracy: 0.970 in 12.77s ( 1197 docs/s)
          Perceptron classifier : 	 15296 train docs (  1916 positive)    469 test docs (    49 positive) accuracy: 0.949 in 12.77s ( 1197 docs/s)
                 SGD classifier : 	 15296 train docs (  1916 positive)    469 test docs (    49 positive) accuracy: 0.964 in 12.77s ( 1197 docs/s)
      NB Multinomial classifier : 	 15296 train docs (  1916 positive)    469 test docs (    49 positive) accuracy: 0.940 in 12.80s ( 1195 docs/s)


  Passive-Aggressive classifier : 	 17708 train docs (  2180 positive)    469 test docs (    49 positive) accuracy: 0.938 in 14.86s ( 1191 docs/s)
          Perceptron classifier : 	 17708 train docs (  2180 positive)    469 test docs (    49 positive) accuracy: 0.968 in 14.86s ( 1191 docs/s)
                 SGD classifier : 	 17708 train docs (  2180 positive)    469 test docs (    49 positive) accuracy: 0.968 in 14.86s ( 1191 docs/s)
      NB Multinomial classifier : 	 17708 train docs (  2180 positive)    469 test docs (    49 positive) accuracy: 0.945 in 14.89s ( 1189 docs/s)


Plot results

In [6]:
plot_acc=[]
colors = ['blue', 'green', 'red', 'cyan', 'magenta', 'yellow']
fig = tools.make_subplots(rows=2, cols=2,
                          subplot_titles=(
                          'Classification accuracy as a function of<br>training examples (#)',
                          'Classification accuracy as a function of<br>runtime (s) ' ,
                          'Training Times ',
                          'Prediction Times (%d instances)' % n_test_documents))

def plot_accuracy(x, y,colors, name, legend):
    """Plot accuracy as a function of x."""
    x = np.array(x)
    y = np.array(y)
    trace=go.Scatter(x=x, y=y, mode="lines",
                     name=name,showlegend=legend,
                     line=dict(color=colors))
    plot_acc.append(trace)

rcParams['legend.fontsize'] = 10
cls_names = list(sorted(cls_stats.keys()))
j=0

for _, stats in sorted(cls_stats.items()):
    # Plot accuracy evolution with #examples
    accuracy, n_examples = zip(*stats['accuracy_history'])
    plot_accuracy(n_examples, accuracy,colors[j],
                 cls_names[j],True)
    j=j+1

i=0

for _, stats in sorted(cls_stats.items()):
    # Plot accuracy evolution with runtime
    accuracy, runtime = zip(*stats['runtime_history'])
    plot_accuracy(runtime, accuracy, colors[i],
                 cls_names[i],False)
    i=i+1

for k in range(0,4):
    fig.append_trace(plot_acc[k], 1, 1)

for k in range(4,8):
    fig.append_trace(plot_acc[k], 1, 2) 
    
fig['layout']['xaxis1'].update(title='Training Examples (#)')
fig['layout']['yaxis1'].update(title='Accuracy',
                              range=[0.8,1])

fig['layout']['xaxis2'].update(title='Runtime (s)')
fig['layout']['yaxis2'].update(title='Accuracy',
                               range=[0.8,1])

# Plot fitting times
cls_runtime = []

for cls_name, stats in sorted(cls_stats.items()):
    cls_runtime.append(stats['total_fit_time'])

cls_runtime.append(total_vect_time)
cls_names.append('Vectorization')

rectangles = go.Bar(y=cls_runtime,x=cls_names,
                    showlegend=False,
                    marker=dict(color=colors)
                   )
fig.append_trace(rectangles, 2, 1)
fig['layout']['yaxis3'].update(title='runtime (s)')

# Plot prediction times
cls_runtime = []
cls_names = list(sorted(cls_stats.keys()))
for cls_name, stats in sorted(cls_stats.items()):
    cls_runtime.append(stats['prediction_time'])
cls_runtime.append(parsing_time)
cls_names.append('Read/Parse<br>+Feat.Extr.')
cls_runtime.append(vectorizing_time)
cls_names.append('Hashing<br>+Vect.')

rectangles1 = go.Bar(y= cls_runtime, x=cls_names,
                     showlegend=False,
                     marker=dict(color=colors)
                    )

fig.append_trace(rectangles1, 2, 2)
fig['layout']['yaxis4'].update(title='runtime (s)')


fig['layout'].update(height=900)
This is the format of your plot grid:
[ (1,1) x1,y1 ]  [ (1,2) x2,y2 ]
[ (2,1) x3,y3 ]  [ (2,2) x4,y4 ]

In [7]:
py.iplot(fig, validate=False)
Out[7]:

License

Authors:

    Eustache Diemert <eustache@diemert.fr>
    FedericoV <https://github.com/FedericoV/>

License:

    BSD 3 clause
Still need help?
Contact Us

For guaranteed 24 hour response turnarounds, upgrade to a Developer Support Plan.