为什么我需要在TensorFlow中输入数据的字典数据类型

Question

(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()
X_train = X_train.astype(np.float32).reshape(-1, 28*28) / 255.0
X_test = X_test.astype(np.float32).reshape(-1, 28*28) / 255.0
y_train = y_train.astype(np.int32)
y_test = y_test.astype(np.int32)
X_valid, X_train = X_train[:5000], X_train[5000:]
y_valid, y_train = y_train[:5000], y_train[5000:]

dnn_clf = tf.estimator.DNNClassifier(hidden_units=[300,100], n_classes=10,
feature_columns=feature_cols)

input_fn = tf.estimator.inputs.numpy_input_fn(
x={"X": X_train}, y=y_train, num_epochs=40, batch_size=50, shuffle=True)

dnn_clf.train(input_fn=input_fn)

My question is: What's the difference between

x={"X": X_train}, y=y_train, num_epochs=40, batch_size=50, shuffle=True)

and

x=X_train, y=y_train, num_epochs=40, batch_size=50, shuffle=True)

Why can't I just use a list X_train for the input x , but have to convert X_train to a dictionary x={"X": X_train}

Why I don't need to do the same thing for y ?

Answer 1

According to documentation TensorFlow 1.11 for numpy_input_fn , you can use both approaches. But I've tested it for TensorFlow 1.9 and it didn't work. So, it seems that this functionality was added later.

If I understood, another question was "why x parameter is a map and y is just an array". The thing is that model can have multiple input feature columns/tensors, but the output tensor can be only one. That's true for, at least, standard estimators and for general use-cases. So, you need to somehow specify multiple inputs and the most simple way is just to pass a map.