Linear models with CNN features

In [1]:
# Rather than importing everything manually, we'll make things easy
#   and load them all in, and just import them from there.
%matplotlib inline
import utils3; 
from utils3 import *
Using Theano backend.
/home/karel/anaconda3/lib/python3.6/site-packages/theano/gpuarray/ 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 6020 on context None
Mapped name None to device cuda: Graphics Device (0000:02:00.0)


We need to find a way to convert the imagenet predictions to a probability of being a cat or a dog, since that is what the Kaggle competition requires us to submit. We could use the imagenet hierarchy to download a list of all the imagenet categories in each of the dog and cat groups, and could then solve our problem in various ways, such as:

  • Finding the largest probability that's either a cat or a dog, and using that label
  • Averaging the probability of all the cat categories and comparing it to the average of all the dog categories.

But these approaches have some downsides:

  • They require manual coding for something that we should be able to learn from the data
  • They ignore information available in the predictions; for instance, if the models predicts that there is a bone in the image, it's more likely to be a dog than a cat.

A very simple solution to both of these problems is to learn a linear model that is trained using the 1,000 predictions from the imagenet model for each image as input, and the dog/cat label as target.

In [2]:
%matplotlib inline
import os, json
from glob import glob
import numpy as np
import scipy
from sklearn.preprocessing import OneHotEncoder
from sklearn.metrics import confusion_matrix
np.set_printoptions(precision=4, linewidth=100)
from matplotlib import pyplot as plt
import utils3
from utils3 import plots, get_batches, plot_confusion_matrix, get_data
In [3]:
from numpy.random import random, permutation
from scipy import misc, ndimage
from scipy.ndimage.interpolation import zoom

import keras
from keras import backend as K
from keras.utils.data_utils import get_file
from keras.models import Sequential
from keras.layers import Input
from keras.layers.core import Flatten, Dense, Dropout, Lambda
from keras.layers.convolutional import Conv2D, MaxPooling2D, ZeroPadding2D
from keras.optimizers import SGD, RMSprop
from keras.preprocessing import image

Linear models in keras

It turns out that each of the Dense() layers is just a linear model, followed by a simple activation function. We'll learn about the activation function later - first, let's review how linear models work.

A linear model is (as I'm sure you know) simply a model where each row is calculated as sum(row * weights), where weights needs to be learnt from the data, and will be the same for every row. For example, let's create some data that we know is linearly related:

In [6]:
x = random((30,2))
y =, [2., 3.]) + 1.
In [7]:
array([[ 0.0106,  0.363 ],
       [ 0.478 ,  0.3925],
       [ 0.7236,  0.3616],
       [ 0.1891,  0.4651],
       [ 0.4663,  0.7481]])
In [8]:
array([ 2.1102,  3.1335,  3.532 ,  2.7734,  4.1768])

We can use keras to create a simple linear model (Dense() - with no activation - in Keras) and optimize it using SGD to minimize mean squared error (mse):

In [9]:
lm = Sequential([ Dense(1, input_shape=(2,)) ])
lm.compile(optimizer=SGD(lr=0.1), loss='mse')

(See the Optim Tutorial notebook and associated Excel spreadsheet to learn all about SGD and related optimization algorithms.)

This has now learnt internal weights inside the lm model, which we can use to evaluate the loss function (MSE).

In [10]:
lm.evaluate(x, y, verbose=0)
In [11]:, y, epochs=5, batch_size=1)
Epoch 1/5
30/30 [==============================] - 0s - loss: 0.7089      
Epoch 2/5
30/30 [==============================] - 0s - loss: 0.0475     
Epoch 3/5
30/30 [==============================] - 0s - loss: 0.0224     
Epoch 4/5
30/30 [==============================] - 0s - loss: 0.0094     
Epoch 5/5
30/30 [==============================] - 0s - loss: 0.0042     
<keras.callbacks.History at 0x7f9a7fcb51d0>
In [12]:
lm.evaluate(x, y, verbose=0)

And, of course, we can also take a look at the weights - after fitting, we should see that they are close to the weights we used to calculate y (2.0, 3.0, and 1.0).

In [13]:
[array([[ 2.0453],
        [ 2.7755]], dtype=float32), array([ 1.0656], dtype=float32)]

Train linear model on predictions

Using a Dense() layer in this way, we can easily convert the 1,000 predictions given by our model into a probability of dog vs cat--simply train a linear model to take the 1,000 predictions as input, and return dog or cat as output, learning from the Kaggle data. This should be easier and more accurate than manually creating a map from imagenet categories to one dog/cat category.

Training the model

We start with some basic config steps. We copy a small amount of our data into a 'sample' directory, with the exact same structure as our 'train' directory--this is always a good idea in all machine learning, since we should do all of our initial testing using a dataset small enough that we never have to wait for it.

In [4]:
#path = "data/dogscats/sample/"
path = "data/dogscats/"
model_path = path + 'models/'
if not os.path.exists(model_path): os.mkdir(model_path)

We will process as many images at a time as our graphics card allows. This is a case of trial and error to find the max batch size - the largest size that doesn't give an out of memory error.

In [5]:

We need to start with our VGG 16 model, since we'll be using its predictions and features.

In [6]:
from vgg16_3 import Vgg16
vgg = Vgg16()
model = vgg.model

Our overall approach here will be:

  1. Get the true labels for every image
  2. Get the 1,000 imagenet category predictions for every image
  3. Feed these predictions as input to a simple linear model.

Let's start by grabbing training and validation batches.

In [7]:
# Use batch size of 1 since we're just doing preprocessing on the CPU
val_batches = get_batches(path+'valid', shuffle=False, batch_size=1)
batches = get_batches(path+'train', shuffle=False, batch_size=1)
Found 2000 images belonging to 2 classes.
Found 23000 images belonging to 2 classes.

Loading and resizing the images every time we want to use them isn't necessary - instead we should save the processed arrays. By far the fastest way to save and load numpy arrays is using bcolz. This also compresses the arrays, so we save disk space. Here are the functions we'll use to save and load using bcolz.

In [8]:
import bcolz
def save_array(fname, arr): c=bcolz.carray(arr, rootdir=fname, mode='w'); c.flush()
def load_array(fname): return[:]

We have provided a simple function that joins the arrays from all the batches - let's use this to grab the training and validation data:

In [9]:
val_data = get_data(path+'valid')
Found 2000 images belonging to 2 classes.
In [10]:
trn_data = get_data(path+'train')
Found 23000 images belonging to 2 classes.
In [11]:
(23000, 3, 224, 224)
In [12]:
save_array(model_path+ 'train_data.bc', trn_data)
save_array(model_path + 'valid_data.bc', val_data)

We can load our training and validation data later without recalculating them:

In [8]:
trn_data = load_array(model_path+'train_data.bc')
val_data = load_array(model_path+'valid_data.bc')
In [13]:
(2000, 3, 224, 224)
In [14]:

Keras returns classes as a single column, so we convert to one hot encoding

In [9]:
def onehot(x): return np.array(OneHotEncoder().fit_transform(x.reshape(-1,1)).todense())
In [10]:
val_classes = val_batches.classes
trn_classes = batches.classes
val_labels = onehot(val_classes)
trn_labels = onehot(trn_classes)
In [11]:
(23000, 2)
In [18]:
array([0, 0, 0, 0], dtype=int32)
In [19]:
array([[ 1.,  0.],
       [ 1.,  0.],
       [ 1.,  0.],
       [ 1.,  0.]])

...and their 1,000 imagenet probabilties from VGG16--these will be the features for our linear model:

In [20]:
trn_features = model.predict(trn_data, batch_size=batch_size)
val_features = model.predict(val_data, batch_size=batch_size)
In [20]:
(23000, 1000)
In [21]:
save_array(model_path+ 'train_lastlayer_features.bc', trn_features)
save_array(model_path + 'valid_lastlayer_features.bc', val_features)

We can load our training and validation features later without recalculating them:

In [12]:
trn_features = load_array(model_path+'train_lastlayer_features.bc')
val_features = load_array(model_path+'valid_lastlayer_features.bc')

Now we can define our linear model, just like we did earlier:

In [13]:
# 1000 inputs, since that's the saved features, and 2 outputs, for dog and cat
lm = Sequential([ Dense(2, activation='softmax', input_shape=(1000,)) ])
lm.compile(optimizer=RMSprop(lr=0.1), loss='categorical_crossentropy', metrics=['accuracy'])

We're ready to fit the model!

In [14]:
In [15]:
In [16]:, trn_labels, epochs=3, batch_size=batch_size, 
       validation_data=(val_features, val_labels))
Train on 23000 samples, validate on 2000 samples
Epoch 1/3
23000/23000 [==============================] - 2s - loss: 0.1342 - acc: 0.9703 - val_loss: 0.1537 - val_acc: 0.9720
Epoch 2/3
23000/23000 [==============================] - 2s - loss: 0.1638 - acc: 0.9741 - val_loss: 0.1795 - val_acc: 0.9740
Epoch 3/3
23000/23000 [==============================] - 2s - loss: 0.1830 - acc: 0.9745 - val_loss: 0.1986 - val_acc: 0.9755
<keras.callbacks.History at 0x7f63be4909e8>
In [27]:
Layer (type)                 Output Shape              Param #   
dense_4 (Dense)              (None, 2)                 2002      
Total params: 2,002.0
Trainable params: 2,002
Non-trainable params: 0.0

Viewing model prediction examples

Keras' fit() function conveniently shows us the value of the loss function, and the accuracy, after every epoch ("epoch" refers to one full run through all training examples). The most important metrics for us to look at are for the validation set, since we want to check for over-fitting.

  • Tip: with our first model we should try to overfit before we start worrying about how to handle that - there's no point even thinking about regularization, data augmentation, etc if you're still under-fitting! (We'll be looking at these techniques shortly).

As well as looking at the overall metrics, it's also a good idea to look at examples of each of:

  1. A few correct labels at random
  2. A few incorrect labels at random
  3. The most correct labels of each class (ie those with highest probability that are correct)
  4. The most incorrect labels of each class (ie those with highest probability that are incorrect)
  5. The most uncertain labels (ie those with probability closest to 0.5).

Let's see what we, if anything, we can from these (in general, these are particularly useful for debugging problems in the model; since this model is so simple, there may not be too much to learn at this stage.)

Calculate predictions on validation set, so we can find correct and incorrect examples:

In [28]:
# We want both the classes...
preds = lm.predict_classes(val_features, batch_size=batch_size)
# ...and the probabilities of being a cat
probs = lm.predict_proba(val_features, batch_size=batch_size)[:,0]
  64/2000 [..............................] - ETA: 0s
array([ 1.,  1.,  1.,  1.,  1.,  1.,  1.,  1.], dtype=float32)
In [29]:
array([0, 0, 0, 0, 0, 0, 0, 0])

Get the filenames for the validation set, so we can view images:

In [30]:
filenames = val_batches.filenames
In [31]:
# Number of images to view for each visualization task
n_view = 4

Helper function to plot images by index in the validation set:

In [32]:
def plots_idx(idx, titles=None):
    plots([image.load_img(path + 'valid/' + filenames[i]) for i in idx], titles=titles)
In [33]:
#1. A few correct labels at random
correct = np.where(preds==val_labels[:,1])[0]
idx = permutation(correct)[:n_view]
plots_idx(idx, probs[idx])
In [34]:
#2. A few incorrect labels at random
incorrect = np.where(preds!=val_labels[:,1])[0]
idx = permutation(incorrect)[:n_view]
plots_idx(idx, probs[idx])
In [35]:
#3. The images we most confident were cats, and are actually cats
correct_cats = np.where((preds==0) & (preds==val_labels[:,1]))[0]
most_correct_cats = np.argsort(probs[correct_cats])[::-1][:n_view]
plots_idx(correct_cats[most_correct_cats], probs[correct_cats][most_correct_cats])
In [36]:
# as above, but dogs
correct_dogs = np.where((preds==1) & (preds==val_labels[:,1]))[0]
most_correct_dogs = np.argsort(probs[correct_dogs])[:n_view]
plots_idx(correct_dogs[most_correct_dogs], 1-probs[correct_dogs][most_correct_dogs])
In [37]:
#3. The images we were most confident were cats, but are actually dogs
incorrect_cats = np.where((preds==0) & (preds!=val_labels[:,1]))[0]
most_incorrect_cats = np.argsort(probs[incorrect_cats])[::-1][:n_view]
plots_idx(incorrect_cats[most_incorrect_cats], probs[incorrect_cats][most_incorrect_cats])