basic example using tf estimators as of 1.4

tf_estimator_example.py

def input_fn(total_repeats=int(1e8),batch_size=100):
    x=tf.constant([
        [[0,1]],
        [[0,0]],
        [[1,1]],
        [[1,0]]
    ],dtype=tf.float32)


    y=tf.constant([
        [[1]],
        [[0]],
        [[0]],
        [[1]]
    ],dtype=tf.int32)
    data=tf.data.Dataset.from_tensor_slices((x,y))
    data=data.repeat(total_repeats)
    data=data.batch(batch_size)
    it=data.make_one_shot_iterator()
    x,y=it.get_next()

    return x,y

def val_fn(total_repeats=1,batch_size=4):
    x=tf.constant([
        [[0,1]],
        [[0,0]],
        [[1,1]],
        [[1,0]]
    ],dtype=tf.float32)


    y=tf.constant([
        [[1]],
        [[0]],
        [[0]],
        [[1]]
    ],dtype=tf.int32)
    data=tf.data.Dataset.from_tensor_slices((x,y))
    data=data.repeat(total_repeats)
    data=data.batch(batch_size)
    it=data.make_one_shot_iterator()
    x,y=it.get_next()

    return x,y

def model_fn(features,labels,mode,params):
    net=features
    print(mode)
    if mode == tf.estimator.ModeKeys.TRAIN:
        is_training=True
    else:
        is_training=False
    normalizer_params={'is_training': is_training, 'decay': 0.9, 'updates_collections': None}
    with sm.arg_scope([sm.fully_connected, sm.conv2d],
                  normalizer_fn=sm.batch_norm,
                  normalizer_params= normalizer_params,
                  weights_regularizer=sm.l2_regularizer(float(1e-04)),
                  weights_initializer=layers.xavier_initializer(),
                  activation_fn=sm.nn.relu
                  ) as asc: pass
    with sm.arg_scope(asc):
        net=sm.fully_connected(net,24)
        net=sm.fully_connected(net,24)
    yhat=sm.fully_connected(net,1)
    
    global_step=tf.train.get_or_create_global_step()
    adam=None
    metrics=None
    loss=None
    if is_training or mode==tf.estimator.ModeKeys.EVAL:
        loss=tf.losses.mean_squared_error(tf.cast(labels,dtype=tf.float32),tf.cast(yhat,dtype=tf.float32))
        loss=tf.losses.get_total_loss()
        adam=tf.train.AdamOptimizer(learning_rate=.01)
        adam=adam.minimize(loss,global_step=global_step)
        metrics={'mae':tf.metrics.mean_absolute_error(tf.cast(labels,dtype=tf.float32),tf.cast(yhat,dtype=tf.float32))}

    return tf.estimator.EstimatorSpec(mode=mode,predictions=yhat,loss=loss,train_op=adam,eval_metric_ops=metrics)

esto=tf.estimator.Estimator(model_fn)
esto.train(input_fn,max_steps=10000)
esto.evaluate(input_fn,steps=1)
list(esto.predict(val_fn,))