In [2]:

    

%matplotlib inline


    

In [3]:

    

path = "data/dogscats-redux/"


    

In [4]:

    

from __future__ import division,print_function

import os, json
from glob import glob
import numpy as np
np.set_printoptions(precision=4, linewidth=100)
from matplotlib import pyplot as plt


    

In [5]:

    

import utils; reload(utils)
from utils import plots


    

    




Using gpu device 0: Tesla K80 (CNMeM is disabled, cuDNN 5103)
/home/ubuntu/anaconda2/lib/python2.7/site-packages/theano/sandbox/cuda/__init__.py:600: UserWarning: Your cuDNN version is more recent than the one Theano officially supports. If you see any problems, try updating Theano or downgrading cuDNN to version 5.
  warnings.warn(warn)
Using Theano backend.

In [6]:

    

batch_size=64


    

In [7]:

    

# Import our class, and instantiate
import vgg16; reload(vgg16)
from vgg16 import Vgg16


    

In [8]:

    

vgg = Vgg16()


    

In [9]:

    

batches = vgg.get_batches(path+'train', batch_size=4)


    

    




Found 22776 images belonging to 2 classes.

In [10]:

    

imgs,labels = next(batches)
plots(imgs, titles=labels)


    

In [11]:

    

vgg.predict(imgs, True)


    

    
Out[11]:





(array([ 0.5245,  0.1812,  0.1423,  0.9877], dtype=float32),
 array([207, 281, 332, 284]),
 [u'golden_retriever', u'tabby', u'Angora', u'Siamese_cat'])

In [12]:

    

vgg.finetune(batches)


    

In [13]:

    

val_batches = vgg.get_batches(path+'valid', batch_size=batch_size)


    

    




Found 2200 images belonging to 2 classes.

In [14]:

    

vgg.fit(batches, val_batches, nb_epoch=1)


    

    




Epoch 1/1
22776/22776 [==============================] - 664s - loss: 0.2304 - acc: 0.9705 - val_loss: 0.1941 - val_acc: 0.9805

In [15]:

    

vgg.predict(imgs, True)


    

    
Out[15]:





(array([ 1.,  1.,  1.,  1.], dtype=float32),
 array([1, 0, 0, 0]),
 ['dogs', 'cats', 'cats', 'cats'])

In [32]:

    

batches = vgg.get_batches(path+'train', batch_size=32, shuffle=False)
imgs,labels = next(batches)
results = vgg.predict(imgs, True)


    

    




Found 22776 images belonging to 2 classes.

In [33]:

    

plots(imgs[0:4], titles=labels[0:4])


    

In [41]:

    

batches, preds = vgg.test(path+'test', batch_size = 32)


    

    




Found 12500 images belonging to 1 classes.

In [44]:

    

#Save our test results arrays so we can use them again later
filenames = batches.filenames
np.save(path + 'test_preds.dat', preds)
np.save(path + 'filenames.dat', filenames)


    

In [49]:

    

preds


    

    
Out[49]:





array([[  1.0000e+00,   1.2281e-36],
       [  9.9999e-01,   1.0085e-05],
       [  1.0000e+00,   3.2340e-37],
       ..., 
       [  1.0000e+00,   2.2421e-44],
       [  1.0000e+00,   1.4013e-45],
       [  4.5341e-20,   1.0000e+00]], dtype=float32)

In [59]:

    

batches.filenames
ids = np.array([int(f[7:f.find('.')]) for f in filenames])
print(ids[:5])
print(filenames[:5])


    

    




[ 9292 12026  9688  4392   779]
['unkown/9292.jpg', 'unkown/12026.jpg', 'unkown/9688.jpg', 'unkown/4392.jpg', 'unkown/779.jpg']

In [63]:

    

isdog = preds[:,1]


    

In [64]:

    

subm = np.stack([ids,isdog], axis=1)
subm[:5]


    

    
Out[64]:





array([[  9.2920e+03,   1.2281e-36],
       [  1.2026e+04,   1.0085e-05],
       [  9.6880e+03,   3.2340e-37],
       [  4.3920e+03,   2.8026e-45],
       [  7.7900e+02,   1.0000e+00]])

In [65]:

    

np.savetxt(path+'submission.csv', subm, fmt='%d,%.5f', header='id,label', comments='')


    

In [ ]: