在 DNNClassifier 中将 list 或 ndarray 作为 feature_column 传递

Question

import tensorflow as tf
import pandas as pd


a = [[1,1],[2,2],[3,3]]
b = [11,22,33]

mydata = pd.DataFrame({'images':a,'labels':b})
feature_columns = [tf.feature_column.numeric_column('images',shape=[1,1])]


train_input_fn = tf.estimator.inputs.pandas_input_fn(x =mydata,
                                                 y=mydata['labels'],
                                                 batch_size=60,
                                                 num_epochs=1,
                                                 shuffle=True)


estimator = tf.estimator.DNNClassifier(hidden_units=[64,32,16],
                                   feature_columns=feature_columns,
                                   n_classes=2)


estimator.train(input_fn=train_input_fn,steps=100)

我得到的错误是

INFO:tensorflow:Create CheckpointSaverHook.
INFO:tensorflow:Error reported to Coordinator: <class 'tensorflow.python.framework.errors_impl.InternalError'>, Unable to get element as bytes.
INFO:tensorflow:Saving checkpoints for 0 into /tmp/tmptver1w_k/model.ckpt.

TypeError                                 Traceback (most recent call last)
TypeError: expected bytes, list found

阅读多个stackoverflow页面和github ...它必须与saving_listeners做一些事情。但无法弄清楚。

请帮忙。

Answer 1

TF Estimator 期望字节作为 x 的输入。

试试这个，它应该可以解决这个错误：

a = [bytes([1,1]), bytes([2,2]), bytes([3,3])]