notebook.community

Edit and run 





In [1]:



    


import graphlab
graphlab.canvas.set_target('ipynb')

Deep Learning Basics

Today we'll walk through building a deep learning network for image recognition CIFAR-10 dataset. The CIFAR data-set represents real-world data that is already formatted and labeled, so we can focus on building our network today instead of cleaning the data.

We're going to walk through 4 steps today:

  • Loading the Data
  • Training our Model
  • Evaluating the results of the model
  • Look at ways to improveing the performance of our model

We've download a subset of the CIFAR-10 data set for you. We load the data into an SFame, which is a powerful and scalable data structure that is used by many of the models in GraphLab Create.





In [2]:



    


image_train = graphlab.SFrame('https://static.turi.com/datasets/coursera/deep_learning/image_train_data')
image_test = graphlab.SFrame('https://static.turi.com/datasets/coursera/deep_learning/image_test_data')



    


















    






[INFO] This commercial license of GraphLab Create is assigned to engr@turi.com.

[INFO] Start server at: ipc:///tmp/graphlab_server-98991 - Server binary: /Users/piotrteterwak/.virtualenvs/gl-1.4/lib/python2.7/site-packages/graphlab/unity_server - Server log: /tmp/graphlab_server_1442964667.log
[INFO] GraphLab Server Version: 1.6










    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_train_data/dir_archive.ini to /var/tmp/graphlab-piotrteterwak/98991/51c6a729-39ce-497e-a6e4-32b47158b5be.ini










    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_train_data/objects.bin to /var/tmp/graphlab-piotrteterwak/98991/99e89616-5a60-4315-82f5-a198f20ee633.bin










    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_train_data/m_8b0619884fada761.frame_idx to /var/tmp/graphlab-piotrteterwak/98991/bb1e963e-9ee5-4a5e-a477-682948e3ea88.frame_idx










    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_train_data/m_8b0619884fada761.sidx to /var/tmp/graphlab-piotrteterwak/98991/8b59f77c-29da-4ff9-b925-94e8615e73f5.sidx










    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_test_data/dir_archive.ini to /var/tmp/graphlab-piotrteterwak/98991/4034beac-c3da-4b74-9400-a532da514a1e.ini










    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_test_data/objects.bin to /var/tmp/graphlab-piotrteterwak/98991/ac7cfe1b-5671-4d87-9f40-9e9130c126a8.bin










    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_test_data/m_686d0b1bd5620cb1.frame_idx to /var/tmp/graphlab-piotrteterwak/98991/6b43b4c2-9a6e-45f7-99c2-0a4a449178cf.frame_idx










    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_test_data/m_686d0b1bd5620cb1.sidx to /var/tmp/graphlab-piotrteterwak/98991/8184f530-f346-452d-a09b-6dbe8d3f1454.sidx

Exploring the data





In [3]:



    


image_train



    


















    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_train_data/m_8b0619884fada761.0000 to /var/tmp/graphlab-piotrteterwak/98991/a25f18f8-7a29-4ee4-9559-c13719f94172.0000










    
Out[3]:








    

        id
        image
        label
        deep_features
        image_array
    

    

        24
        Height: 32 Width: 32
        bird
        [0.242871761322,
1.09545373917, 0.0, ...		
        [73.0, 77.0, 58.0, 71.0,
68.0, 50.0, 77.0, 69.0, ...		
    

    

        33
        Height: 32 Width: 32
        cat
        [0.525087952614, 0.0,
0.0, 0.0, 0.0, 0.0, ...		
        [7.0, 5.0, 8.0, 7.0, 5.0,
8.0, 5.0, 4.0, 6.0, 7.0, ...		
    

    

        36
        Height: 32 Width: 32
        cat
        [0.566015958786, 0.0,
0.0, 0.0, 0.0, 0.0, ...		
        [169.0, 122.0, 65.0,
131.0, 108.0, 75.0, ...		
    

    

        70
        Height: 32 Width: 32
        dog
        [1.12979578972, 0.0, 0.0,
0.778194487095, 0.0, ...		
        [154.0, 179.0, 152.0,
159.0, 183.0, 157.0, ...		
    

    

        90
        Height: 32 Width: 32
        bird
        [1.71786928177, 0.0, 0.0,
0.0, 0.0, 0.0, ...		
        [216.0, 195.0, 180.0,
201.0, 178.0, 160.0, ...		
    

    

        97
        Height: 32 Width: 32
        automobile
        [1.57818555832, 0.0, 0.0,
0.0, 0.0, 0.0, ...		
        [33.0, 44.0, 27.0, 29.0,
44.0, 31.0, 32.0, 45.0, ...		
    

    

        107
        Height: 32 Width: 32
        dog
        [0.0, 0.0,
0.220677852631, 0.0,  ...		
        [97.0, 51.0, 31.0, 104.0,
58.0, 38.0, 107.0, 61.0, ...		
    

    

        121
        Height: 32 Width: 32
        bird
        [0.0, 0.23753464222, 0.0,
0.0, 0.0, 0.0, ...		
        [93.0, 96.0, 88.0, 102.0,
106.0, 97.0, 117.0, ...		
    

    

        136
        Height: 32 Width: 32
        automobile
        [0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 7.5737862587, 0.0, ...		
        [35.0, 59.0, 53.0, 36.0,
56.0, 56.0, 42.0, 62.0, ...		
    

    

        138
        Height: 32 Width: 32
        bird
        [0.658935725689, 0.0,
0.0, 0.0, 0.0, 0.0, ...		
        [205.0, 193.0, 195.0,
200.0, 187.0, 193.0, ...		
    



[2005 rows x 5 columns]
Note: Only the head of the SFrame is printed.
You can use print_rows(num_rows=m, num_columns=n) to print more rows and columns.



















In [4]:



    


graphlab.image_analysis.resize(image_train['image'], 128, 128, 3).show()

We now use the neuralnet_classifier provided by GraphLab Create to create a neural network for our data set. The create method picks a default network architecture for you based on the dataset.





In [5]:



    


raw_pixel_model = graphlab.logistic_classifier.create(image_train, features=['image_array'], target='label')



    


















    







PROGRESS: Logistic regression:










    







PROGRESS: --------------------------------------------------------










    







PROGRESS: Number of examples          : 1902










    







PROGRESS: Number of classes           : 4










    







PROGRESS: Number of feature columns   : 1










    







PROGRESS: Number of unpacked features : 3072










    







PROGRESS: Number of coefficients    : 9219










    







PROGRESS: Starting L-BFGS










    







PROGRESS: --------------------------------------------------------










    







PROGRESS: +-----------+----------+-----------+--------------+-------------------+---------------------+










    







PROGRESS: | Iteration | Passes   | Step size | Elapsed Time | Training-accuracy | Validation-accuracy |










    







PROGRESS: +-----------+----------+-----------+--------------+-------------------+---------------------+










    







PROGRESS: | 1         | 6        | 0.000010  | 3.915201     | 0.305994          | 0.310680            |










    







PROGRESS: | 2         | 8        | 1.000000  | 5.005213     | 0.380652          | 0.310680            |










    







PROGRESS: | 3         | 9        | 1.000000  | 5.642424     | 0.404837          | 0.388350            |










    







PROGRESS: | 4         | 10       | 1.000000  | 6.282546     | 0.413249          | 0.368932            |










    







PROGRESS: | 5         | 11       | 1.000000  | 6.935667     | 0.420610          | 0.398058            |










    







PROGRESS: | 6         | 12       | 1.000000  | 7.559418     | 0.364879          | 0.291262            |










    







PROGRESS: | 10        | 17       | 1.000000  | 10.457879    | 0.497371          | 0.436893            |










    







PROGRESS: +-----------+----------+-----------+--------------+-------------------+---------------------+










    






PROGRESS: Creating a validation set from 5 percent of training data. This may take a while.
          You can set ``validation_set=None`` to disable validation tracking.

In order to ensure that the deep learning model is acutally learning how to recognize the data, instead of memorizing features, we want to validate it with a dataset it hasn't seen before.

Making a single prediction





In [6]:



    


graphlab.image_analysis.resize(image_test[0:1]['image'], 128, 128, 3).show()



    


















    







PROGRESS: Downloading https://static.turi.com/datasets/coursera/deep_learning/image_test_data/m_686d0b1bd5620cb1.0000 to /var/tmp/graphlab-piotrteterwak/98991/729e4e3f-ce1f-4319-a2fd-b57f4a6eea08.0000










    



























In [7]:



    


image_test[0:1]['label']



    


















    
Out[7]:








dtype: str
Rows: 1
['cat']

















In [8]:



    


raw_pixel_model.predict(image_test[0:1])



    


















    
Out[8]:








dtype: str
Rows: 1
['dog']

Evaluating the model





In [31]:



    


raw_pixel_model.evaluate(image_test)



    


















    
Out[31]:








{'accuracy': 0.46225, 'confusion_matrix': Columns:
 	target_label	str
 	predicted_label	str
 	count	int
 
 Rows: 16
 
 Data:
 +--------------+-----------------+-------+
 | target_label | predicted_label | count |
 +--------------+-----------------+-------+
 |     dog      |       cat       |  149  |
 |  automobile  |       dog       |  190  |
 |     bird     |       dog       |  259  |
 |  automobile  |       cat       |   75  |
 |  automobile  |    automobile   |  602  |
 |     bird     |    automobile   |  106  |
 |     dog      |    automobile   |   93  |
 |     cat      |       dog       |  380  |
 |     dog      |       dog       |  485  |
 |     cat      |       cat       |  217  |
 +--------------+-----------------+-------+
 [16 rows x 3 columns]
 Note: Only the head of the SFrame is printed.
 You can use print_rows(num_rows=m, num_columns=n) to print more rows and columns.}

Our network does well on the data, but we'd like to do better. How can we improve it?





In [10]:



    


# if you wanted to do this on your own
# step 1: deep_learning_model = graphlab.load_model('imagenet_model/')
# step 2: image_train['deep_features'] = deep_learning_model.extract_features(image_train)



    











In [11]:



    


deep_model = graphlab.neuralnet_classifier.create(image_train, features=['image'], target='label')



    


















    







PROGRESS: Computing mean image...










    







PROGRESS: Done computing mean image.










    







PROGRESS: Creating neuralnet using cpu










    







PROGRESS: Training with batch size = 100










    







PROGRESS: +-----------+----------+--------------+-------------------+---------------------+-----------------+










    







PROGRESS: | Iteration | Examples | Elapsed Time | Training-accuracy | Validation-accuracy | Examples/second |










    







PROGRESS: +-----------+----------+--------------+-------------------+---------------------+-----------------+










    







PROGRESS: | 1         | 2000     | 1.567609     | 0.305000          | 0.378641            | 1275.828613     |










    







PROGRESS: | 2         | 2000     | 2.797574     | 0.385500          | 0.504854            | 1626.391235     |










    







PROGRESS: | 3         | 2000     | 3.969070     | 0.447500          | 0.592233            | 1707.442139     |










    







PROGRESS: | 4         | 2000     | 5.235676     | 0.505500          | 0.553398            | 1579.210205     |










    







PROGRESS: | 5         | 2000     | 6.448314     | 0.549000          | 0.592233            | 1649.498657     |










    







PROGRESS: | 6         | 2000     | 7.728839     | 0.567000          | 0.601942            | 1562.195190     |










    







PROGRESS: | 7         | 2000     | 9.101758     | 0.574500          | 0.631068            | 1456.916870     |










    







PROGRESS: | 8         | 2000     | 10.345062    | 0.599000          | 0.660194            | 1608.810547     |










    







PROGRESS: | 9         | 2000     | 11.642864    | 0.623000          | 0.621359            | 1541.266968     |










    







PROGRESS: | 10        | 2000     | 12.812695    | 0.630000          | 0.650485            | 1709.868774     |

Evaluating the Improved Model





In [17]:



    


graphlab.image_analysis.resize(image_test[0:1]['image'], 128, 128, 3).show()



    


















    



























In [18]:



    


image_test[0:1]['label']



    


















    
Out[18]:








dtype: str
Rows: 1
['cat']

















In [19]:



    


deep_model.predict(image_test[0:1])



    


















    
Out[19]:








dtype: str
Rows: 1
['cat']

















In [32]:



    


deep_model.evaluate(image_test)



    


















    
Out[32]:








{'accuracy': 0.597000002861023, 'confusion_matrix': Columns:
 	target_label	str
 	predicted_label	str
 	count	int
 
 Rows: 16
 
 Data:
 +--------------+-----------------+-------+
 | target_label | predicted_label | count |
 +--------------+-----------------+-------+
 |     bird     |       bird      |  511  |
 |     cat      |       bird      |  104  |
 |     dog      |       bird      |  134  |
 |  automobile  |       bird      |   25  |
 |     bird     |       cat       |  186  |
 |     cat      |       cat       |  521  |
 |     dog      |       cat       |  320  |
 |  automobile  |       cat       |   56  |
 |     bird     |       dog       |  164  |
 |     cat      |       dog       |  244  |
 +--------------+-----------------+-------+
 [16 rows x 3 columns]
 Note: Only the head of the SFrame is printed.
 You can use print_rows(num_rows=m, num_columns=n) to print more rows and columns.}

Improving the model further with Deep Features





In [21]:



    


deep_features_model = graphlab.logistic_classifier.create(image_train, features=['deep_features'], target='label')



    


















    







PROGRESS: WARNING: Detected extremely low variance for feature(s) 'deep_features' because all entries are nearly the same.
Proceeding with model training using all features. If the model does not provide results of adequate quality, exclude the above mentioned feature(s) from the input dataset.










    







PROGRESS: Logistic regression:










    







PROGRESS: --------------------------------------------------------










    







PROGRESS: Number of examples          : 1903










    







PROGRESS: Number of classes           : 4










    







PROGRESS: Number of feature columns   : 1










    







PROGRESS: Number of unpacked features : 4096










    







PROGRESS: Number of coefficients    : 12291










    







PROGRESS: Starting L-BFGS










    







PROGRESS: --------------------------------------------------------










    







PROGRESS: +-----------+----------+-----------+--------------+-------------------+---------------------+










    







PROGRESS: | Iteration | Passes   | Step size | Elapsed Time | Training-accuracy | Validation-accuracy |










    







PROGRESS: +-----------+----------+-----------+--------------+-------------------+---------------------+










    







PROGRESS: | 1         | 5        | 0.000131  | 4.282864     | 0.743037          | 0.715686            |










    







PROGRESS: | 2         | 9        | 0.250000  | 7.641871     | 0.768786          | 0.735294            |










    







PROGRESS: | 3         | 10       | 0.250000  | 8.691117     | 0.771939          | 0.735294            |










    







PROGRESS: | 4         | 11       | 0.250000  | 9.755411     | 0.778245          | 0.754902            |










    







PROGRESS: | 5         | 12       | 0.250000  | 10.764233    | 0.784025          | 0.745098            |










    







PROGRESS: | 6         | 13       | 0.250000  | 11.726264    | 0.793484          | 0.774510            |










    







PROGRESS: | 7         | 14       | 0.250000  | 12.695556    | 0.818707          | 0.803922            |










    







PROGRESS: | 8         | 15       | 0.250000  | 13.687111    | 0.843405          | 0.784314            |










    







PROGRESS: | 9         | 16       | 0.250000  | 14.631095    | 0.856017          | 0.803922            |










    







PROGRESS: | 10        | 17       | 0.250000  | 15.605435    | 0.865476          | 0.803922            |










    







PROGRESS: +-----------+----------+-----------+--------------+-------------------+---------------------+










    






PROGRESS: Creating a validation set from 5 percent of training data. This may take a while.
          You can set ``validation_set=None`` to disable validation tracking.

Evaluating the deep model





In [33]:



    


deep_features_model.evaluate(image_test)



    


















    
Out[33]:








{'accuracy': 0.784, 'confusion_matrix': Columns:
 	target_label	str
 	predicted_label	str
 	count	int
 
 Rows: 16
 
 Data:
 +--------------+-----------------+-------+
 | target_label | predicted_label | count |
 +--------------+-----------------+-------+
 |  automobile  |       cat       |   14  |
 |     dog      |       cat       |  232  |
 |  automobile  |       dog       |   5   |
 |     cat      |       bird      |   69  |
 |     bird     |       dog       |   58  |
 |     dog      |       bird      |   43  |
 |     cat      |    automobile   |   38  |
 |     bird     |       cat       |  133  |
 |     dog      |    automobile   |   21  |
 |     dog      |       dog       |  704  |
 +--------------+-----------------+-------+
 [16 rows x 3 columns]
 Note: Only the head of the SFrame is printed.
 You can use print_rows(num_rows=m, num_columns=n) to print more rows and columns.}

Finding Similar Images





In [23]:



    


nearest_neighbors_model = graphlab.nearest_neighbors.create(image_train, features=['deep_features'],
                                                            label='id')



    


















    







PROGRESS: Starting brute force nearest neighbors model training.

















In [24]:



    


def get_nearest_neighbors(image):
    ans = nearest_neighbors_model.query(image)
    return image_train.filter_by(ans['reference_label'],'id')

Similar Images: Cat Photo





In [26]:



    


cat = image_train[18:19]
graphlab.image_analysis.resize(cat['image'],128,128,3).show()



    


















    



























In [28]:



    


graphlab.image_analysis.resize(get_nearest_neighbors(cat)['image'],128,128,3).show()



    


















    







PROGRESS: Starting pairwise querying.










    







PROGRESS: +--------------+---------+-------------+--------------+










    







PROGRESS: | Query points | # Pairs | % Complete. | Elapsed Time |










    







PROGRESS: +--------------+---------+-------------+--------------+










    







PROGRESS: | 0            | 1       | 0.0498753   | 27.056ms     |










    







PROGRESS: | Done         |         | 100         | 350.123ms    |










    







PROGRESS: +--------------+---------+-------------+--------------+

Similar Images: Car Photo





In [29]:



    


car = image_train[8:9]
graphlab.image_analysis.resize(car['image'], 128, 128, 3).show()



    


















    



























In [30]:



    


graphlab.image_analysis.resize(get_nearest_neighbors(car)['image'],256,256,3).show()



    


















    







PROGRESS: Starting pairwise querying.










    







PROGRESS: +--------------+---------+-------------+--------------+










    







PROGRESS: | Query points | # Pairs | % Complete. | Elapsed Time |










    







PROGRESS: +--------------+---------+-------------+--------------+










    







PROGRESS: | 0            | 1       | 0.0498753   | 13.922ms     |










    







PROGRESS: | Done         |         | 100         | 395.811ms    |










    







PROGRESS: +--------------+---------+-------------+--------------+










    



























In [ ]:

Content source: turi-code/tutorials

Similar notebooks:

notebook.community | gallery | about