diff --git a/03_classification.ipynb b/03_classification.ipynb
index ae2ee96..ea39159 100644
--- a/03_classification.ipynb
+++ b/03_classification.ipynb
@@ -460,7 +460,9 @@
   {
    "cell_type": "code",
    "execution_count": 34,
-   "metadata": {},
+   "metadata": {
+    "collapsed": true
+   },
    "outputs": [],
    "source": [
     "y_scores = cross_val_predict(sgd_clf, X_train, y_train_5, cv=3,\n",
@@ -2061,19 +2063,660 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "## 4. Spam classifier\n",
-    "\n",
-    "Coming soon..."
+    "## 4. Spam classifier"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "First, let's fetch the data:"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 126,
    "metadata": {
     "collapsed": true
    },
    "outputs": [],
-   "source": []
+   "source": [
+    "import os\n",
+    "import tarfile\n",
+    "from six.moves import urllib\n",
+    "\n",
+    "DOWNLOAD_ROOT = \"http://spamassassin.apache.org/old/publiccorpus/\"\n",
+    "HAM_URL = DOWNLOAD_ROOT + \"20030228_easy_ham.tar.bz2\"\n",
+    "SPAM_URL = DOWNLOAD_ROOT + \"20030228_spam.tar.bz2\"\n",
+    "SPAM_PATH = os.path.join(\"datasets\", \"spam\")\n",
+    "\n",
+    "def fetch_spam_data(spam_url=SPAM_URL, spam_path=SPAM_PATH):\n",
+    "    if not os.path.isdir(spam_path):\n",
+    "        os.makedirs(spam_path)\n",
+    "    for filename, url in ((\"ham.tar.bz2\", HAM_URL), (\"spam.tar.bz2\", SPAM_URL)):\n",
+    "        path = os.path.join(spam_path, filename)\n",
+    "        if not os.path.isfile(path):\n",
+    "            urllib.request.urlretrieve(url, path)\n",
+    "        tar_bz2_file = tarfile.open(path)\n",
+    "        tar_bz2_file.extractall(path=SPAM_PATH)\n",
+    "        tar_bz2_file.close()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 127,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "fetch_spam_data()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Next, let's load all the emails:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 128,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "HAM_DIR = os.path.join(SPAM_PATH, \"easy_ham\")\n",
+    "SPAM_DIR = os.path.join(SPAM_PATH, \"spam\")\n",
+    "ham_filenames = [name for name in sorted(os.listdir(HAM_DIR)) if len(name) > 20]\n",
+    "spam_filenames = [name for name in sorted(os.listdir(SPAM_DIR)) if len(name) > 20]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 129,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "len(ham_filenames)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 130,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "len(spam_filenames)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We can use Python's `email` module to parse these emails (this handles headers, encoding, and so on):"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 131,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "import email\n",
+    "import email.policy\n",
+    "\n",
+    "def load_email(is_spam, filename, spam_path=SPAM_PATH):\n",
+    "    directory = \"spam\" if is_spam else \"easy_ham\"\n",
+    "    with open(os.path.join(spam_path, directory, filename), \"rb\") as f:\n",
+    "        return email.parser.BytesParser(policy=email.policy.default).parse(f)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 132,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "ham_emails = [load_email(is_spam=False, filename=name) for name in ham_filenames]\n",
+    "spam_emails = [load_email(is_spam=True, filename=name) for name in spam_filenames]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Let's look at one example of ham and one example of spam, to get a feel of what the data looks like:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 133,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print(ham_emails[1].get_content().strip())"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 134,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print(spam_emails[6].get_content().strip())"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Some emails are actually multipart, with images and attachments (which can have their own attachments). Let's look at the various types of structures we have:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 135,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "def get_email_structure(email):\n",
+    "    if isinstance(email, str):\n",
+    "        return email\n",
+    "    payload = email.get_payload()\n",
+    "    if isinstance(payload, list):\n",
+    "        return \"multipart({})\".format(\", \".join([\n",
+    "            get_email_structure(sub_email)\n",
+    "            for sub_email in payload\n",
+    "        ]))\n",
+    "    else:\n",
+    "        return email.get_content_type()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 136,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "from collections import Counter\n",
+    "\n",
+    "def structures_counter(emails):\n",
+    "    structures = Counter()\n",
+    "    for email in emails:\n",
+    "        structure = get_email_structure(email)\n",
+    "        structures[structure] += 1\n",
+    "    return structures"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 137,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "structures_counter(ham_emails).most_common()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 138,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "structures_counter(spam_emails).most_common()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "It seems that the ham emails are more often plain text, while spam has quite a lot of HTML. Moreover, quite a few ham emails are signed using PGP, while no spam is. In short, it seems that the email structure is a usual information to have."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Now let's take a look at the email headers:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 139,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "for header, value in spam_emails[0].items():\n",
+    "    print(header,\":\",value)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "There's probably a lot of useful information in there, such as the sender's email address (12a1mailbot1@web.de looks fishy), but we will just focus on the `Subject` header:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 140,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "spam_emails[0][\"Subject\"]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Okay, before we learn too much about the data, let's not forget to split it into a training set and a test set:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 141,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "from sklearn.model_selection import train_test_split\n",
+    "\n",
+    "X = np.array(ham_emails + spam_emails)\n",
+    "y = np.array([0] * len(ham_emails) + [1] * len(spam_emails))\n",
+    "\n",
+    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Okay, let's start writing the preprocessing functions. First, we will need a function to convert HTML to plain text. Arguably the best way to do this would be to use the great [BeautifulSoup](https://www.crummy.com/software/BeautifulSoup/) library, but I would like to avoid adding another dependency to this project, so let's hack a quick & dirty solution using regular expressions (at the risk of [un̨ho͞ly radiańcé destro҉ying all enli̍̈́̂̈́ghtenment](https://stackoverflow.com/a/1732454/38626)). The following function first drops the `<head>` section, then converts all `<a>` tags to the word HYPERLINK, then it gets rid of all HTML tags, leaving only the plain text. For readability, it also replaces multiple newlines with single newlines, and finally it unescapes html entities (such as `&gt;` or `&nbsp;`):"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 142,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "import re\n",
+    "from html import unescape\n",
+    "\n",
+    "def html_to_plain_text(html):\n",
+    "    text = re.sub('<head.*?>.*?</head>', '', html, flags=re.M | re.S | re.I)\n",
+    "    text = re.sub('<a\\s.*?>', ' HYPERLINK ', text, flags=re.M | re.S | re.I)\n",
+    "    text = re.sub('<.*?>', '', text, flags=re.M | re.S)\n",
+    "    text = re.sub(r'(\\s*\\n)+', '\\n', text, flags=re.M | re.S)\n",
+    "    return unescape(text)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Let's see if it works. This is HTML spam:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 143,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "html_spam_emails = [email for email in X_train[y_train==1]\n",
+    "                    if get_email_structure(email) == \"text/html\"]\n",
+    "sample_html_spam = html_spam_emails[7]\n",
+    "print(sample_html_spam.get_content().strip()[:1000], \"...\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "And this is the resulting plain text:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 144,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print(html_to_plain_text(sample_html_spam.get_content())[:1000], \"...\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Great! Now let's write a function that takes an email as input and returns its content as plain text, whatever its format is:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 145,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "def email_to_text(email):\n",
+    "    html = None\n",
+    "    for part in email.walk():\n",
+    "        ctype = part.get_content_type()\n",
+    "        if not ctype in (\"text/plain\", \"text/html\"):\n",
+    "            continue\n",
+    "        try:\n",
+    "            content = part.get_content()\n",
+    "        except: # in case of encoding issues\n",
+    "            content = str(part.get_payload())\n",
+    "        if ctype == \"text/plain\":\n",
+    "            return content\n",
+    "        else:\n",
+    "            html = content\n",
+    "    if html:\n",
+    "        return html_to_plain_text(html)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 146,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print(email_to_text(sample_html_spam)[:100], \"...\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Let's throw in some stemming! For this to work, you need to install the Natural Language Toolkit ([NLTK](http://www.nltk.org/)). It's as simple as running the following command (don't forget to activate your virtualenv first; if you don't have one, you will likely need administrator rights, or use the `--user` option):\n",
+    "\n",
+    "`$ pip install nltk`"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 147,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "try:\n",
+    "    import nltk\n",
+    "\n",
+    "    stemmer = nltk.PorterStemmer()\n",
+    "    for word in (\"Computations\", \"Computation\", \"Computing\", \"Computed\", \"Compute\", \"Compulsive\"):\n",
+    "        print(word, \"=>\", stemmer.stem(word))\n",
+    "except ImportError:\n",
+    "    print(\"Error: stemming requires the NLTK module.\")\n",
+    "    stemmer = None"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We will also need a way to replace URLs with the word \"URL\". For this, we could use hard core [regular expressions](https://mathiasbynens.be/demo/url-regex) but we will just use the [urlextract](https://github.com/lipoja/URLExtract) library. You can install it with the following command (don't forget to activate your virtualenv first; if you don't have one, you will likely need administrator rights, or use the `--user` option):\n",
+    "\n",
+    "`$ pip install urlextract`"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 148,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "try:\n",
+    "    import urlextract # may require an Internet connection to download root domain names\n",
+    "    \n",
+    "    url_extractor = urlextract.URLExtract()\n",
+    "    print(url_extractor.find_urls(\"Will it detect github.com and https://youtu.be/7Pq-S557XQU?t=3m32s\"))\n",
+    "except ImportError:\n",
+    "    print(\"Error: replacing URLs requires the urlextract module.\")\n",
+    "    url_extractor = None"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We are ready to put all this together into a transformer that we will use to convert emails to word counters. Note that we split sentences into words using Python's `split()` method, which uses whitespaces for word boundaries. This works for many written languages, but not all. For example, Chinese and Japanese scripts generally don't use spaces between words, and Vietnamese often uses spaces even between syllables. It's okay in this exercise, because the dataset is (mostly) in English."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 149,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "from sklearn.base import BaseEstimator, TransformerMixin\n",
+    "\n",
+    "class EmailToWordCounterTransformer(BaseEstimator, TransformerMixin):\n",
+    "    def __init__(self, strip_headers=True, lower_case=True, remove_punctuation=True,\n",
+    "                 replace_urls=True, replace_numbers=True, stemming=True):\n",
+    "        self.strip_headers = strip_headers\n",
+    "        self.lower_case = lower_case\n",
+    "        self.remove_punctuation = remove_punctuation\n",
+    "        self.replace_urls = replace_urls\n",
+    "        self.replace_numbers = replace_numbers\n",
+    "        self.stemming = stemming\n",
+    "    def fit(self, X, y=None):\n",
+    "        return self\n",
+    "    def transform(self, X, y=None):\n",
+    "        X_transformed = []\n",
+    "        for email in X:\n",
+    "            text = email_to_text(email) or \"\"\n",
+    "            if self.lower_case:\n",
+    "                text = text.lower()\n",
+    "            if self.replace_urls and url_extractor is not None:\n",
+    "                urls = list(set(url_extractor.find_urls(text)))\n",
+    "                urls.sort(key=lambda url: len(url), reverse=True)\n",
+    "                for url in urls:\n",
+    "                    text = text.replace(url, \" URL \")\n",
+    "            if self.replace_numbers:\n",
+    "                text = re.sub(r'\\d+(?:\\.\\d*(?:[eE]\\d+))?', 'NUMBER', text)\n",
+    "            if self.remove_punctuation:\n",
+    "                text = re.sub(r'\\W+', ' ', text, flags=re.M)\n",
+    "            word_counts = Counter(text.split())\n",
+    "            if self.stemming and stemmer is not None:\n",
+    "                stemmed_word_counts = Counter()\n",
+    "                for word, count in word_counts.items():\n",
+    "                    stemmed_word = stemmer.stem(word)\n",
+    "                    stemmed_word_counts[stemmed_word] += count\n",
+    "                word_counts = stemmed_word_counts\n",
+    "            X_transformed.append(word_counts)\n",
+    "        return np.array(X_transformed)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Let's try this transformer on a few emails:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 150,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "X_few = X_train[:3]\n",
+    "X_few_wordcounts = EmailToWordCounterTransformer().fit_transform(X_few)\n",
+    "X_few_wordcounts"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "This looks about right!"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Now we have the word counts, and we need to convert them to vectors. For this, we will build another transformer whose `fit()` method will build the vocabulary (an ordered list of the most common words) and whose `transform()` method will use the vocabulary to convert word counts to vectors. The output is a sparse matrix."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 151,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "from scipy.sparse import csr_matrix\n",
+    "\n",
+    "class WordCounterToVectorTransformer(BaseEstimator, TransformerMixin):\n",
+    "    def __init__(self, vocabulary_size=1000):\n",
+    "        self.vocabulary_size = vocabulary_size\n",
+    "    def fit(self, X, y=None):\n",
+    "        total_count = Counter()\n",
+    "        for word_count in X:\n",
+    "            for word, count in word_count.items():\n",
+    "                total_count[word] += min(count, 10)\n",
+    "        most_common = total_count.most_common()[:self.vocabulary_size]\n",
+    "        self.most_common_ = most_common\n",
+    "        self.vocabulary_ = {word: index + 1 for index, (word, count) in enumerate(most_common)}\n",
+    "        return self\n",
+    "    def transform(self, X, y=None):\n",
+    "        rows = []\n",
+    "        cols = []\n",
+    "        data = []\n",
+    "        for row, word_count in enumerate(X):\n",
+    "            for word, count in word_count.items():\n",
+    "                rows.append(row)\n",
+    "                cols.append(self.vocabulary_.get(word, 0))\n",
+    "                data.append(count)\n",
+    "        return csr_matrix((data, (rows, cols)), shape=(len(X), self.vocabulary_size + 1))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 152,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "vocab_transformer = WordCounterToVectorTransformer(vocabulary_size=10)\n",
+    "X_few_vectors = vocab_transformer.fit_transform(X_few_wordcounts)\n",
+    "X_few_vectors"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 153,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "X_few_vectors.toarray()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "What does this matrix mean? Well, the 65 in the third row, first column, means that the third email contains 65 words that are not part of the vocabulary. The 0 next to it means that the first word in the vocabulary is not present in this email. The 1 next to it means that the second word is present once, and so on. You can look at the vocabulary to know which words we are talking about. The first word is \"the\", the second word is \"of\", etc."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 154,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "vocab_transformer.vocabulary_"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We are now ready to train our first spam classifier! Let's transform the whole dataset:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 155,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "from sklearn.pipeline import Pipeline\n",
+    "\n",
+    "preprocess_pipeline = Pipeline([\n",
+    "    (\"email_to_wordcount\", EmailToWordCounterTransformer()),\n",
+    "    (\"wordcount_to_vector\", WordCounterToVectorTransformer()),\n",
+    "])\n",
+    "\n",
+    "X_train_transformed = preprocess_pipeline.fit_transform(X_train)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 156,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from sklearn.linear_model import LogisticRegression\n",
+    "from sklearn.model_selection import cross_val_score\n",
+    "\n",
+    "log_clf = LogisticRegression()\n",
+    "score = cross_val_score(log_clf, X_train_transformed, y_train, cv=3, verbose=3)\n",
+    "score.mean()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Over 98.7%, not bad for a first try! :) However, remember that we are using the \"easy\" dataset. You can try with the harder datasets, the results won't be so amazing. You would have to try multiple models, select the best ones and fine-tune them using cross-validation, and so on.\n",
+    "\n",
+    "But you get the picture, so let's stop now, and just print out the precision/recall we get on the test set:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 157,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from sklearn.metrics import precision_score, recall_score\n",
+    "\n",
+    "X_test_transformed = preprocess_pipeline.transform(X_test)\n",
+    "\n",
+    "log_clf = LogisticRegression()\n",
+    "log_clf.fit(X_train_transformed, y_train)\n",
+    "\n",
+    "y_pred = log_clf.predict(X_test_transformed)\n",
+    "\n",
+    "print(\"Precision: {:.2f}%\".format(precision_score(y_test, y_pred)))\n",
+    "print(\"Recall: {:.2f}%\".format(recall_score(y_test, y_pred)))"
+   ]
   }
  ],
  "metadata": {