{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Demo 1 - Distanz\n",
"\n",
"Dies ist ein Jupyter Notebook, welches auf mybinder.org gehostet wird. Alle Eingaben werden nach Beendigung gelöscht. \n",
"\n",
"Jupyter Notebooks enthalten Text- oder Codeblöcke. Codeblöcke werden durch SHIFT-ENTER ausgeführt (Dreiecksymbol oben rechts auf Mobil).\n",
"\n",
"Solange ein Block ausgeführt wird, wird nicht mehr die Nr [3] des Blocks, sondern [*] angezeigt.\n",
"\n",
"**<a href=\"./Demo2-Objekterkennung.ipynb\" target=\"_blank\">Link zu Demo 2</a>**"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"##### Initialisierung"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Gensim ist die Standard Word2Vec-Implementation\n",
"import gensim\n",
"\n",
"# Check, welche Version genutzt wird. Ändert sich häufig mit starken Änderungen.\n",
"print(gensim.__version__)\n",
"\n",
"# Modell mit Top 5000 deutschen Worten laden, 6MB\n",
"model = gensim.models.KeyedVectors.load_word2vec_format(\"min5000.model\", binary=True)\n",
"\n",
"# FutureWarning ignorieren\n",
"import warnings\n",
"warnings.filterwarnings('ignore')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Nächste Nachbarn anzeigen"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Die 10 Wörter laden, die München am Nächsten sind und Wahrscheinlichkeiten dazu ausgegen.\n",
"\n",
"matches = model.most_similar(positive=[\"Muenchen\"], negative=None, topn=10)\n",
"\n",
"matches"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Wort-Mathematik"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Muenchen ist zu Bayern wie Hannover zu ???\n",
"\n",
"matches = model.most_similar(positive=[\"Hannover\", \"Bayern\"], negative=[\"Muenchen\"], topn=10)\n",
"\n",
"matches"
]
},
{
"cell_type": "markdown",
"metadata": {
"collapsed": true
},
"source": [
"### Visualisierung von Wordvektoren"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"%matplotlib inline\n",
"import matplotlib.pyplot as plt\n",
"from sklearn.decomposition import PCA\n",
"from sklearn.manifold import TSNE\n",
"\n",
"print('geladen')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Definieren der Anzeigefunktion"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"# By: https://github.com/devmount/GermanWordEmbeddings/blob/master/visualize.py\n",
"\n",
"# function draw_words\n",
"# ... reduces dimensionality of vectors of given words either with PCA or with t-SNE and draws the words into a diagram\n",
"# @param word2vec model to visualize vectors from\n",
"# @param list words list of word strings to visualize\n",
"# @param bool pca use PCA (True) or t-SNE (False) to reduce dimensionality \n",
"# @param bool alternate use different color and label align for every second word\n",
"# @param bool arrows use arrows to connect related words (items that are next to each other in list)\n",
"# @param float x1 x axis range (from)\n",
"# @param float x2 x axis range (to)\n",
"# @param float y1 y axis range (from)\n",
"# @param float y2 y axis range (to)\n",
"# @param string title for diagram\n",
"def draw_words(model, words, pca=False, alternate=True, arrows=True, x1=3, x2=3, y1=3, y2=3, title=''):\n",
" # get vectors for given words from model\n",
" vectors = [model[word] for word in words]\n",
"\n",
" if pca:\n",
" pca = PCA(n_components=2, whiten=True)\n",
" vectors2d = pca.fit(vectors).transform(vectors)\n",
" else:\n",
" tsne = TSNE(n_components=2, random_state=0)\n",
" vectors2d = tsne.fit_transform(vectors)\n",
"\n",
" # draw image\n",
" plt.figure(figsize=(6,6))\n",
" if pca:\n",
" plt.axis([x1, x2, y1, y2])\n",
"\n",
" first = True # color alternation to divide given groups\n",
" for point, word in zip(vectors2d , words):\n",
" # plot points\n",
" plt.scatter(point[0], point[1], c='r' if first else 'g')\n",
" # plot word annotations\n",
" plt.annotate(\n",
" word, \n",
" xy = (point[0], point[1]),\n",
" xytext = (-7, -6) if first else (7, -6),\n",
" textcoords = 'offset points',\n",
" ha = 'right' if first else 'left',\n",
" va = 'bottom',\n",
" size = \"x-large\"\n",
" )\n",
" first = not first if alternate else first\n",
"\n",
" # draw arrows\n",
" if arrows:\n",
" for i in range(0, len(words)-1, 2):\n",
" a = vectors2d[i][0] + 0.04\n",
" b = vectors2d[i][1]\n",
" c = vectors2d[i+1][0] - 0.04\n",
" d = vectors2d[i+1][1]\n",
" plt.arrow(a, b, c-a, d-b,\n",
" shape='full',\n",
" lw=0.1,\n",
" edgecolor='#bbbbbb',\n",
" facecolor='#bbbbbb',\n",
" length_includes_head=True,\n",
" head_width=0.08,\n",
" width=0.01\n",
" )\n",
"\n",
" # draw diagram title\n",
" if title:\n",
" plt.title(title)\n",
"\n",
" plt.tight_layout()\n",
" plt.show()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### 2-D Visualisierung von Nachbarn"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Nachbarn suchen und anzeigen\n",
"matches = model.most_similar(positive=[\"denken\"], negative=[], topn=10)\n",
"words = [match[0] for match in matches]\n",
"draw_words(model, words, True, False, False, -3, 3, -3, 3, r'$PCA\\ Visualisierung:\\ Nachbarn$')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.6.3"
}
},
"nbformat": 4,
"nbformat_minor": 2
}