In [1]:

    

from tensorflow.examples.tutorials.mnist import input_data


    

In [2]:

    

mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)


    

    




Extracting MNIST_data/train-images-idx3-ubyte.gz
Extracting MNIST_data/train-labels-idx1-ubyte.gz
Extracting MNIST_data/t10k-images-idx3-ubyte.gz
Extracting MNIST_data/t10k-labels-idx1-ubyte.gz

In [3]:

    

import deepchem as dc
import tensorflow as tf
from deepchem.models.tensorgraph.layers import Layer, Input, Reshape, Flatten
from deepchem.models.tensorgraph.layers import Dense, SoftMaxCrossEntropy, ReduceMean, SoftMax


    

In [4]:

    

train = dc.data.NumpyDataset(mnist.train.images, mnist.train.labels)
valid = dc.data.NumpyDataset(mnist.validation.images, mnist.validation.labels)


    

In [5]:

    

tg = dc.models.TensorGraph(tensorboard=True, model_dir='/tmp/mnist')
feature = Input(shape=(None, 784))
tg.add_layer(feature)
tg.add_feature(feature)

# Images are square 28x28 (batch, height, width, channel)
make_image = Reshape(shape=(-1, 28, 28, 1))
tg.add_layer(make_image, parents=[feature])

class Conv2d(Layer):
    def __init__(self, num_outputs, kernel_size=5, **kwargs):
        self.num_outputs = num_outputs
        self.kernel_size = kernel_size
        super().__init__(**kwargs)
    def __call__(self, *parents):
        parent_tensor = parents[0].out_tensor
        out_tensor = tf.contrib.layers.conv2d(parent_tensor,
                                              num_outputs=self.num_outputs,
                                              kernel_size = self.kernel_size,
                                              padding="SAME",
                                              activation_fn=tf.nn.relu,
                                              normalizer_fn=tf.contrib.layers.batch_norm)
        self.out_tensor = tf.nn.max_pool(out_tensor, 
                                         ksize=[1, 2, 2, 1],
                                         strides=[1, 2, 2, 1], 
                                         padding='SAME')
        return self.out_tensor
conv2d_1 = Conv2d(num_outputs=32)
tg.add_layer(conv2d_1, parents=[make_image])

conv2d_2 = Conv2d(num_outputs=64)
tg.add_layer(conv2d_2, parents=[conv2d_1])

flatten = Flatten()
tg.add_layer(flatten, parents=[conv2d_2])

dense1 = Dense(out_channels=1024, activation_fn=tf.nn.relu)
tg.add_layer(dense1, parents=[flatten])

dense2 = Dense(out_channels=10)
tg.add_layer(dense2, parents=[dense1])

label = Input(shape=(None, 10))
tg.add_layer(label, parents=list())
tg.add_label(label)

smce = SoftMaxCrossEntropy()
tg.add_layer(smce, parents=[label, dense2])

loss = ReduceMean()
tg.add_layer(loss, parents=[smce])
tg.set_loss(loss)

output = SoftMax()
tg.add_layer(output, parents=[dense2])
tg.add_output(output)


    

In [6]:

    

tg.fit(train, nb_epoch=10)
tg.save()


    

    




Training for 10 epochs
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Ending epoch 0: Average loss 0.198305
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Ending epoch 1: Average loss 0.0536864
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Ending epoch 2: Average loss 0.0425871
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Ending epoch 3: Average loss 0.0336403
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Ending epoch 4: Average loss 0.028042
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Ending epoch 5: Average loss 0.0257894
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Ending epoch 6: Average loss 0.0187417
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Ending epoch 7: Average loss 0.01757
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Ending epoch 8: Average loss 0.0132135
On batch 0
On batch 50
On batch 100
On batch 150
On batch 200
On batch 250
On batch 300
On batch 350
On batch 400
On batch 450
On batch 500
On batch 550
On batch 600
On batch 650
On batch 700
On batch 750
On batch 800
On batch 850
On batch 900
On batch 950
On batch 1000
On batch 1050
Loggin
Ending epoch 9: Average loss 0.0121854
TIMING: model fitting took 77.903 s

In [7]:

    

from sklearn.metrics import roc_curve, auc
import numpy as np

print("Validation")
prediction = np.squeeze(tg.predict_on_batch(valid.X))

fpr = dict()
tpr = dict()
roc_auc = dict()
for i in range(10):
    fpr[i], tpr[i], thresh = roc_curve(valid.y[:, i], prediction[:, i])
    roc_auc[i] = auc(fpr[i], tpr[i])
    print("class %s:auc=%s" % (i, roc_auc[i]))


    

    




Validation
class 0:auc=0.99987578265
class 1:auc=0.999975981083
class 2:auc=0.99987010958
class 3:auc=0.999980197583
class 4:auc=0.999985348139
class 5:auc=0.999889990331
class 6:auc=0.999762644083
class 7:auc=0.999936670072
class 8:auc=0.999936085657
class 9:auc=0.999968609514

In [ ]:

    




    

In [ ]: