我错过了什么吗？ TensorFlow 中的简单分类器输入 function 出错

Question

I've been following along with the freecodecamp tutorial on TensorFlow and I've tried to modify a basic classifier to handle one of my own structured datasets.

I have a training dataset and a testing dataset, each one containing some integers and some strings. I'm trying to predict the value in the allocated column, but it keeps throwing this error when the Classifier.train method is called:

UnimplementedError: Cast string to float is not supported
     [[{{node head/losses/Cast}}]]

During handling of the above exception, another exception occurred:

UnimplementedError                        Traceback (most recent call last)
/usr/local/lib/python3.7/dist-packages/tensorflow/python/client/session.py in _do_call(self, fn, *args)
   1392                     '\nsession_config.graph_options.rewrite_options.'
   1393                     'disable_meta_optimizer = True')
-> 1394       raise type(e)(node_def, op, message)  # pylint: disable=no-value-for-parameter
   1395 
   1396   def _extend_graph(self):

UnimplementedError: Cast string to float is not supported
     [[node head/losses/Cast (defined at /usr/local/lib/python3.7/dist-packages/tensorflow_estimator/python/estimator/head/binary_class_head.py:255) ]]

I've tried converting the dataset so that all of the values are integers or floats, but I keep getting the same error. From what I've been able to figure out, the classifier should be able to operate on different datatypes, so I can't see why that would be the issue unless I need to define them somewhere?

I know that it's reading in the data correctly, because when I use the.head() function it's all properly formatted. I've been stuck on this error for days and I can't figure out what I'm missing. Any help would be greatly appreciated. My code is below.

%tensorflow_version 2.x 

from __future__ import absolute_import, division, print_function, unicode_literals
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from IPython.display import clear_output
from six.moves import urllib
import tensorflow.compat.v2.feature_column as fc
import tensorflow as tf

CSV_COLUMN_NAMES = ['GroupNumber', 'GroupUnit', 'GroupSkill1', 'GroupSkill2', 'GroupSkill3', 'GroupSkill4', 'GroupPreference1', 
                'GroupPreference2', 'GroupPreference3', 'ProjectNumber', 'ProjectUnit', 'ProjectSkill1', 'ProjectSkill2', 'ProjectSkill3', 'ProjectSkill4', 'ProjectPreference1', 'ProjectPreference2', 'ProjectPreference3', 'Allocated']
ALLOCATED = [0, 1]

train = pd.read_csv('https://raw.githubusercontent.com/nickjackson862/machine-learning/main/trainData40_10.csv', names=CSV_COLUMN_NAMES, header=0)
test = pd.read_csv('https://raw.githubusercontent.com/nickjackson862/machine-learning/main/testData40_10.csv', names=CSV_COLUMN_NAMES, header=0)

train_y = train.pop('Allocated')
test_y = test.pop('Allocated')
train.head()


def input_fn(features, labels, training=True, batch_size=100):   
    dataset = tf.data.Dataset.from_tensor_slices((dict(features), labels))    
    if training:
        dataset = dataset.shuffle(10).repeat()    
    return dataset.batch(batch_size)

my_feature_columns = []
for key in train.keys():
    my_feature_columns.append(tf.feature_column.numeric_column(key=key))

classifier = tf.estimator.DNNClassifier(
    feature_columns=my_feature_columns,
    hidden_units=[50, 20],
    n_classes=2)

classifier.train(
    input_fn=lambda: input_fn(train, train_y, training=True),
    steps=100)

eval_result = classifier.evaluate(
    input_fn=lambda: input_fn(test, test_y, training=False))

print('\nTest set accuracy: {accuracy:0.3f}\n'.format(**eval_result))

Answer 1

I found the problem in this line where you are creating your feature columns.

my_feature_columns.append(tf.feature_column.numeric_column(key=key))

You are making every feature a numeric feature, but looking at your dataset several of your fields are strings (that CSV file is public btw, you may want to remedy that).

I've tried converting the dataset so that all of the values are integers or floats, but I keep getting the same error.

I believe you did this incorrectly. I just tried running your code but with all the string type columns removed and it ran successfully with no errors. All I did was add the following lines after reading in the CSVs

train.drop(columns=['GroupSkill1', 'GroupSkill2', 'GroupSkill3', 'GroupSkill4', "ProjectSkill1", "ProjectSkill2", "ProjectSkill3", "ProjectSkill4", ], axis=1, inplace=True)
test.drop(columns=['GroupSkill1', 'GroupSkill2', 'GroupSkill3', 'GroupSkill4', "ProjectSkill1", "ProjectSkill2", "ProjectSkill3", "ProjectSkill4", ], axis=1, inplace=True)

Check out this article for advice for creating feature columns for your non-numeric data: https://www.tensorflow.org/tutorials/structured_data/feature_columns