使用TensorFlow训练NN时发生错误：TypeError：'Series'对象是可变的，因此无法进行散列

Question

是的，我错过了转换目标功能的过程。 我尝试以不同的方式编写代码，并在导入数据集后立即将特征转换为整数。 这里的代码：

import math
import tensorflow as tf
import numpy as np
import pandas as pd
from sklearn import preprocessing
from sklearn import metrics
from tensorflow.python.data import Dataset

FEATURES = ["cap-shape", "cap-surface", "cap-color", "bruises", "odor", "gill-attachment",
                              "gill-spacing", "gill-size", "gill-color", "stalk-shape", "stalk-root",
                              "stalk-surface-above-ring", "stalk-surface-below-ring", "stalk-color-above-ring",
                              "stalk-color-below-ring", "veil-type", "veil-color", "ring-number", "ring-type",
                              "spore-print-color", "population", "habitat"]


def preprocess_features(data):
    selected_features = data[["cap-shape", "cap-surface", "cap-color", "bruises", "odor", "gill-attachment",
                              "gill-spacing", "gill-size", "gill-color", "stalk-shape", "stalk-root",
                              "stalk-surface-above-ring", "stalk-surface-below-ring", "stalk-color-above-ring",
                              "stalk-color-below-ring", "veil-type", "veil-color", "ring-number", "ring-type",
                              "spore-print-color", "population", "habitat"]]
    processed_features = selected_features.copy()
    return processed_features


def preprocess_targets(data):
    output_targets = pd.DataFrame()
    output_targets["class"] = data["class"]
    return output_targets


def construct_feature_columns(features):
    return {tf.feature_column.numeric_column(my_feature) for my_feature in features}


def input_fn(features, targets, batch_size=1, num_epochs=None):
    features = {key: np.array(value) for key, value in dict(features).items()}

    ds = Dataset.from_tensor_slices((features, targets))
    ds = ds.batch(batch_size).repeat(num_epochs)

    features, labels = ds.make_one_shot_iterator().get_next()
    return features, labels


def train_nn_classifier_model(learning_rate, steps, batch_size, hidden_units, training_examples, training_targets,
                              validation_examples, validation_targets):
    global validation_root_mean_squared_error, training_root_mean_squared_error
    periods = 10
    steps_per_period = steps / periods

    my_optimizer = tf.train.GradientDescentOptimizer(learning_rate=learning_rate)
    my_optimizer = tf.contrib.estimator.clip_gradients_by_norm(my_optimizer, 5.0)
    train_dnn_classifier = tf.estimator.DNNClassifier(feature_columns=construct_feature_columns(training_examples),
                                                      hidden_units=hidden_units, optimizer=my_optimizer)

    training_input_fn = lambda: input_fn(training_examples, training_targets, batch_size)
    predict_training_input_fn = lambda: input_fn(training_examples, training_targets, num_epochs=1)
    predict_validation_input_fn = lambda: input_fn(validation_examples, validation_targets, num_epochs=1)

    print "Training model..."
    # Error array
    training_rmse = []
    validation_rmse = []
    for period in xrange(0, periods):
        train_dnn_classifier.train(input_fn=training_input_fn(), steps=steps_per_period)

        training_predictions = train_dnn_classifier.predict(input_fn=predict_training_input_fn)
        training_predictions = np.array([it['predictions'][0] for it in training_predictions])

        validation_predictions = train_dnn_classifier.predict(input_fn=predict_validation_input_fn)
        validation_predictions = np.array([it['predictions'][0] for it in validation_predictions])

        training_root_mean_squared_error = math.sqrt(metrics.mean_squared_error(training_predictions, training_targets))
        validation_root_mean_squared_error = math.sqrt(metrics.mean_squared_error(validation_predictions,
                                                                                  validation_targets))
        print "  period %02d : %0.2f" % (period, training_root_mean_squared_error)
        training_rmse.append(training_root_mean_squared_error)
        validation_rmse.append(validation_root_mean_squared_error)
    print "Model training finished."

    print "Final RMSE (on training data):   {:0.2f}".format(training_root_mean_squared_error)
    print "Final RMSE (on validation data): {:0.2f}".format(validation_root_mean_squared_error)

    return train_dnn_classifier


# Train & test set
data_set = pd.read_csv("mushrooms.csv")
for i in range(len(FEATURES)):
    encoder = preprocessing.LabelEncoder()
    encoder.fit(data_set[FEATURES[i]])
    data_set[FEATURES[i]] = encoder.transform(data_set[FEATURES[i]])

msk = np.random.rand(len(data_set)) < 0.8
train_set = data_set[msk]
test_set = data_set[~msk]

# Prepare training and validation on it
training = preprocess_features(train_set)
validation = preprocess_features(train_set)

training_targets = preprocess_targets(train_set)
validation_targets = preprocess_targets(train_set)

# Prepare test
test = preprocess_features(test_set)
test_targets = preprocess_targets(test_set)

dnn_classifier = train_nn_classifier_model(learning_rate=0.001, steps=100, batch_size=100, hidden_units=[10, 10],
                                           training_examples=training, training_targets=training_targets,
                                           validation_examples=validation, validation_targets=validation_targets)

predict_testing_input_fn = lambda: input_fn(test, test_targets, num_epochs=1)

test_predictions = dnn_classifier.predict(input_fn=predict_testing_input_fn)
test_predictions = np.array([item['predictions'][0] for item in test_predictions])

root_mean_squared_error = math.sqrt(metrics.mean_squared_error(test_predictions, test_targets))

print "Final RMSE (on test data): {0.2f}".format(root_mean_squared_error)

具体部分是

for i in range(len(FEATURES)):
    encoder = preprocessing.LabelEncoder()
    encoder.fit(data_set[FEATURES[i]])
    data_set[FEATURES[i]] = encoder.transform(data_set[FEATURES[i]])

其余的应该起作用，因为我是从Google https://colab.research.google.com/notebooks/mlcc/intro_to_neural_nets.ipynb?utm_source=mlcc&utm_campaign=colab-external&utm_medium=referral&utm_content=introneuralnets-colab&hl=zCrollDv但我收到此错误，但不知道如何解决。 你能再帮我一次吗？

Traceback (most recent call last):
  File "/Users/.../individual_task_3.py", line 112, in <module>
    validation_examples=validation, validation_targets=validation_targets)
  File "/Users/.../individual_task_3.py", line 66, in train_nn_classifier_model
    train_dnn_classifier.train(input_fn=training_input_fn(), steps=steps_per_period)
  File "/Users/.../venv/lib/python2.7/site-packages/tensorflow/python/estimator/estimator.py", line 352, in train
    loss = self._train_model(input_fn, hooks, saving_listeners)
  File "/Users/.../venv/lib/python2.7/site-packages/tensorflow/python/estimator/estimator.py", line 809, in _train_model
    input_fn, model_fn_lib.ModeKeys.TRAIN))
  File "/Users/.../venv/lib/python2.7/site-packages/tensorflow/python/estimator/estimator.py", line 668, in _get_features_and_labels_from_input_fn
    result = self._call_input_fn(input_fn, mode)
  File "/Users/.../venv/lib/python2.7/site-packages/tensorflow/python/estimator/estimator.py", line 751, in _call_input_fn
    input_fn_args = util.fn_args(input_fn)
  File "/Users/.../venv/lib/python2.7/site-packages/tensorflow/python/estimator/util.py", line 53, in fn_args
    args = tf_inspect.getfullargspec(fn).args
  File "/Users/.../venv/lib/python2.7/site-packages/tensorflow/python/util/tf_inspect.py", line 65, in getfullargspec
    if d.decorator_argspec is not None), spec_fn(target))
  File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/inspect.py", line 816, in getargspec
    raise TypeError('{!r} is not a Python function'.format(func))
TypeError: ({'habitat': <tf.Tensor 'IteratorGetNext:8' shape=(?,) dtype=int64>, 'cap-surface': <tf.Tensor 'IteratorGetNext:3' shape=(?,) dtype=int64>, 'cap-shape': <tf.Tensor 'IteratorGetNext:2' shape=(?,) dtype=int64>, 'cap-color': <tf.Tensor 'IteratorGetNext:1' shape=(?,) dtype=int64>, 'stalk-color-above-ring': <tf.Tensor 'IteratorGetNext:14' shape=(?,) dtype=int64>, 'stalk-shape': <tf.Tensor 'IteratorGetNext:17' shape=(?,) dtype=int64>, 'ring-number': <tf.Tensor 'IteratorGetNext:11' shape=(?,) dtype=int64>, 'odor': <tf.Tensor 'IteratorGetNext:9' shape=(?,) dtype=int64>, 'gill-attachment': <tf.Tensor 'IteratorGetNext:4' shape=(?,) dtype=int64>, 'stalk-color-below-ring': <tf.Tensor 'IteratorGetNext:15' shape=(?,) dtype=int64>, 'stalk-surface-below-ring': <tf.Tensor 'IteratorGetNext:19' shape=(?,) dtype=int64>, 'gill-spacing': <tf.Tensor 'IteratorGetNext:7' shape=(?,) dtype=int64>, 'spore-print-color': <tf.Tensor 'IteratorGetNext:13' shape=(?,) dtype=int64>, 'gill-color': <tf.Tensor 'IteratorGetNext:5' shape=(?,) dtype=int64>, 'population': <tf.Tensor 'IteratorGetNext:10' shape=(?,) dtype=int64>, 'veil-color': <tf.Tensor 'IteratorGetNext:20' shape=(?,) dtype=int64>, 'ring-type': <tf.Tensor 'IteratorGetNext:12' shape=(?,) dtype=int64>, 'bruises': <tf.Tensor 'IteratorGetNext:0' shape=(?,) dtype=int64>, 'stalk-surface-above-ring': <tf.Tensor 'IteratorGetNext:18' shape=(?,) dtype=int64>, 'veil-type': <tf.Tensor 'IteratorGetNext:21' shape=(?,) dtype=int64>, 'stalk-root': <tf.Tensor 'IteratorGetNext:16' shape=(?,) dtype=int64>, 'gill-size': <tf.Tensor 'IteratorGetNext:6' shape=(?,) dtype=int64>}, <tf.Tensor 'IteratorGetNext:22' shape=(?, 1) dtype=string>) is not a Python function

Answer 1

问题是您正在尝试创建一个集合（使用集合语法{...} ），其元素是panda.Series，这些元素不可哈希，因此不能放入集合中或用作字典的键和tensorflow使用列名作为字典中的键。

似乎如果传递字符串作为功能列名称而不是传递Series对象，则代码应该起作用。