Wayne Nixalo - 12 June 2017

Practical Deep Learning I

Lesson 5: NLP

wordvectors.ipynb code along



In [1]:

    
%matplotlib inline

import os, sys
sys.path.insert(1, os.path.join('utils'))

import utils; reload(utils)
from utils import *
from __future__ import division, print_function









    



/home/wnixalo/miniconda3/envs/FAI/lib/python2.7/site-packages/theano/gpuarray/dnn.py:135: UserWarning: Your cuDNN version is more recent than Theano. If you encounter problems, try updating Theano or downgrading cuDNN to version 5.1.
  warnings.warn("Your cuDNN version is more recent than "
Using cuDNN version 6021 on context None
Mapped name None to device cuda: GeForce GTX 870M (0000:01:00.0)
Using Theano backend.



In [5]:

    
path = 'data/glove.6B/'
res_path = path + 'results/'
if not os.path.exists(path): os.mkdir(path)
if not os.path.exists(res_path): os.mkdir(res_path)

Preprocessing

This section shows how we processed the original glove text files. However, there's no need for you to do this, since we provide the pre-processed glove data --- Link broken; instead I'm downloading from nlp.stanford.edu/projects/glove/glove.6B.zip



In [6]:

    
def get_glove(name):
    with open(path+ 'glove.' + name + '.txt', 'r') as f: lines = [line.split() for line in f]
    words = [d[0] for d in lines]
    vecs = np.stack(np.array(d[1:], dtype=np.float32) for d in lines)
    wordidx = {o:i for i,o in enumerate(words)}
    save_array(res_path+name+'.dat', vecs)
    pickle.dump(words, open(res_path+name+'_words.pkl','wb'))
    pickle.dump(wordidx, open(res_path+name+'_idx.pkl','wb'))



In [7]:

    
get_glove('6B.50d')
# get_glove('6B.100d')
# get_glove('6B.200d')
# get_glove('6B.300d')

Looking at the vectors

After you've downloaded the pre-processed glove data, you should use tar -zxf to untar them, and put them in the path that (res_path) points to.

Then the following functino will return the word vectors as a matrix, the word list, and the mapping from word to index.



In [8]:

    
def load_glove(loc):
    return (load_array(loc + '.dat'),
        pickle.load(open(loc + '_words.pkl', 'rb')),
        pickle.load(open(loc + '_idx.pkl', 'rb')))



In [9]:

    
vecs, words, wirdidx = load_glove(res_path + '6B.50d')
vecs.shape









    Out[9]:





(400000, 50)

Here's the first 25 "words" in glove.



In [10]:

    
' '.join(words[:25])









    Out[10]:





'the , . of to and in a " \'s for - that on is was said with he as it by at ( )'

This is how you can look up a word vector.



In [11]:

    
def w2v(w): return vecs[wirdidx[w]] # mispelled 'wordidx' up above... lol



In [12]:

    
w2v('of')









    Out[12]:





array([ 0.7085,  0.5709, -0.4716,  0.1805,  0.5445,  0.726 ,  0.1816, -0.5239,  0.1038, -0.1757,
        0.0789, -0.3622, -0.1183, -0.8334,  0.1192, -0.1661,  0.0616, -0.0127, -0.5662,  0.0136,
        0.2285, -0.144 , -0.0675, -0.3816, -0.237 , -1.7037, -0.8669, -0.267 , -0.2589,  0.1767,
        3.8676, -0.1613, -0.1327, -0.6888,  0.1844,  0.0052, -0.3387, -0.079 ,  0.2419,  0.3658,
       -0.3473,  0.2848,  0.0757, -0.0622, -0.3899,  0.229 , -0.2162, -0.2256, -0.0939, -0.8037], dtype=float32)

Just for fun, let's take a look at a 2d projection of the first 350 words, using T-SNE



In [13]:

    
reload(sys)
sys.setdefaultencoding('utf8')



In [14]:

    
tsne = TSNE(n_components=2, random_state=0)
Y = tsne.fit_transform(vecs[:500])

start=0; end=350
dat = Y[start:end]
plt.figure(figsize=(15,15))
plt.scatter(dat[:, 0], dat[:, 1])
for label, x, y in zip(words[start:end], dat[:, 0], dat[:, 1]):
    plt.text(x,y,label, color=np.random.rand(3)*0.7,
                 fontsize=14)
plt.show()



In [ ]: