mnist_example.py

from Layers.model import SoftDecisionTree
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

if __name__ == '__main__':

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

    n_features = 784
    n_classes = 10
    batch_size = 32
    val_batch_size = 256


    tree = SoftDecisionTree(max_depth=6,n_features=n_features,n_classes=n_classes,max_leafs=None)
    tree.build_tree()

    # optimizer
    optimizer = tf.train.AdamOptimizer(learning_rate=0.001,beta1=0.9,beta2=0.999,epsilon=1e-08).minimize(tree.loss)

    # Saving the model
    #saver = tf.train.Saver()

    # Initialize the variables (i.e. assign their default value)
    init = tf.global_variables_initializer()

    EPOCHS = 10000
    TOTAL_BATCH = 16
    display_step = 100
    with tf.Session() as sess:
        sess.run(init)
        for epoch in range(EPOCHS):

            avg_cost = 0.
            # Loop over all batches
            acc =0.0
            val_acc = 0.0
            for i in range(TOTAL_BATCH):
                batch_xs, batch_ys = mnist.train.next_batch(batch_size)

                c = tree.boost(X=batch_xs,y=batch_ys,sess=sess)


                target = np.argmax(batch_ys,axis=1)
                preds = tree.predict(X=batch_xs,y=batch_ys,sess=sess)
                acc += accuracy_score(y_pred=preds,y_true=target)/TOTAL_BATCH

                # Compute average loss

                avg_cost+= acc/TOTAL_BATCH
            # Display logs per epoch step
            if (epoch + 1) % display_step == 0:
                batch_val_xs, batch_val_ys = mnist.validation.next_batch(val_batch_size)


                val_target = np.argmax(batch_val_ys, axis=1)
                val_preds = tree.predict(X=batch_val_xs,y=batch_val_ys,sess=sess)
                val_acc = accuracy_score(y_pred=val_preds, y_true=val_target)
                print("Epoch:", '%04d' % (epoch + 1), "cost=",
                      "{:.9f}".format(avg_cost),"training_accuracy=","{:.4f}".format(acc),
                      "validation_accuracy=","{:.4f}".format(val_acc)  )
                #print(collections.Counter(np.argmax(path_probs,axis=1)))

                #print(confusion_matrix(y_true=val_target,y_pred=val_preds) )