Scikit Flow - Digits example

digits.py

import random
from sklearn import datasets, cross_validation, metrics
import tensorflow as tf
from tensorflow.contrib import layers
from tensorflow.contrib import learn

random.seed(42)

# Load dataset and split it into train / test subsets.

digits = datasets.load_digits()
X = digits.images
y = digits.target

X_train, X_test, y_train, y_test = cross_validation.train_test_split(X, y,
    test_size=0.2, random_state=42)

# TensorFlow model using Scikit Flow ops

def conv_model(features, target):
    target = tf.one_hot(target, 10, 1.0, 0.0)
    features = tf.expand_dims(features, 3)
    features = tf.reduce_max(layers.conv2d(features, 12, [3, 3]), [1, 2])
    features = tf.reshape(features, [-1, 12])
    prediction, loss = learn.models.logistic_regression(features, target)
    train_op = layers.optimize_loss(loss,
        tf.contrib.framework.get_global_step(), optimizer='SGD',
        learning_rate=0.01)
    return tf.argmax(prediction, dimension=1), loss, train_op

# Create a classifier, train and predict.
classifier = learn.Estimator(model_fn=conv_model)
classifier.fit(X_train, y_train, steps=1000, batch_size=128)
score = metrics.accuracy_score(classifier.predict(X_test), y_test)
print('Accuracy: %f' % score)

Hey, thanks for this! :)

if im not wrong, line 30 should be:

score = metrics.accuracy_score( y_test,classifier.predict(X_test))

Also, can you explain line 21? :)

@Waffleboy You are right - I'm always flipping y_hat and y_pred arguments.
I just updated code, but previous line 21 meant 2d convolution over input image (e.g. computes weighted average for each square of size 3x3) and then find max across dimensions 1 and 2 - so will be left with [batch_size,12] where for each example in the batch 12 numbers are maximum values of the average from before.

You can think of it as running 12 object detector (or 3x3 objects) across the image and then using their maximum value to determine if object was present or not.