从 tf.data.Datasets 构建 tf,estimator.DNNClassifier

Question

I am new to tensorflow in ML, and thought I could build the model from tf.data.Datasets directly. Here is my code, could not figure out why it did not work. Can someone please advise if it's possible to make it work?

import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
import tensorflow_datasets as tfds

#load the data
train_data, ds_info = tfds.load('mnist', split='train'
                       , shuffle_files=True,with_info=True, as_supervised=True)

feature_columns = [tf.feature_column.numeric_column('x',shape=[28,28])]

#build the model
estimator = tf.estimator.DNNClassifier(
feature_columns=feature_columns,
hidden_units=[300,100],
n_classes=10,
model_dir='/train/DNN')

#train the model
estimator.train(input_fn=train_data)

Answer 1

Please refer working code Build tf.estimator.DNNClassifier from Mnist Dataset.

import tensorflow as tf
import numpy as np
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
learn = tf.contrib.learn
tf.logging.set_verbosity(tf.logging.ERROR)
print(tf.__version__)
##import the dataset
mnist = learn.datasets.load_dataset('mnist')
data = mnist.train.images
labels = np.asarray(mnist.train.labels, dtype=np.int32)
test_data = mnist.test.images
test_labels = np.asarray(mnist.test.labels, dtype = np.int32)
def input(dataset):
    return dataset.images, dataset.labels.astype(np.int32)

# Specify feature
feature_columns = [tf.feature_column.numeric_column(""x"", shape=[28, 28])]
# Build 2 layer DNN classifier
classifier = tf.estimator.DNNClassifier(
    feature_columns=feature_columns,
    hidden_units=[256, 32],
    optimizer=tf.train.AdamOptimizer(1e-4),
    n_classes=10,
    dropout=0.1,
    model_dir=""./tmp/mnist_model""
)

# Define the training inputs
train_input_fn = tf.estimator.inputs.numpy_input_fn(
    x={""x"": input(mnist.train)[0]},
    y=input(mnist.train)[1],
    num_epochs=None,
    batch_size=50,
    shuffle=True
)

classifier.train(input_fn=train_input_fn, steps=100)
# Evaluate accuracy
accuracy_score = classifier.evaluate(input_fn=train_input_fn)[""accuracy""]
print(""\nTrain Accuracy: {0:f}%\n"".format(accuracy_score*100))

# Define the test inputs
test_input_fn = tf.estimator.inputs.numpy_input_fn(
    x={""x"": input(mnist.test)[0]},
    y=input(mnist.test)[1],
    num_epochs=1,
    shuffle=False
)

# Evaluate accuracy
accuracy_score = classifier.evaluate(input_fn=test_input_fn)[""accuracy""]
print(""\nTest Accuracy: {0:f}%\n"".format(accuracy_score*100))"