Keras编程--2 错误处理

1.用SimpleRNN构建分类模型

代码：

# encoding: utf-8

import pandas as pd
import numpy as np
import tensorflow as tf
from keras.preprocessing.text import Tokenizer
# from keras.preprocessing.sequence import pad_sequences #新版中没有这个包
from keras_preprocessing.sequence import pad_sequences
from keras.models import Sequential
# from keras.layers.embeddings import Embedding
from tensorflow.python.keras.layers.embeddings import Embedding
from keras.layers import Dense
from keras.layers import SimpleRNN


# 去掉 方in医案 的频次<=freq_num的医案
def filter_yian(df, freq_num):
    y_train_lst = df["规范方剂"].values # 获取规范方剂的列

    formula_lst, counts_lst = np.unique(y_train_lst, return_counts=True) # 统计y_train_lst中每一个方剂名对应的次数
    # 打印只出现一次的方
    small_formla = []
    for k, v in dict(zip(formula_lst, counts_lst)).items():
        if v <= freq_num:
            # print(k)
            small_formla.append(k)
    print("准备删除的方数：",len(small_formla))

    for index, row in df.iterrows():
        if str(row["规范方剂"]) in small_formla or pd.isna(row["规范化后症状"]):
            df.drop(index=index, axis=0, inplace=True)

    print("剩余医案数量：",df.shape[0])
    print("剩余方剂数量：",len(list(set(df["规范方剂"].values))))
    return df

def get_train_data(df:pd.DataFrame, max_symptoms_length):
    '''

    :param df: 原始数据
    :param dictionary_size:词典大小
    :param max_symptoms_length:设定症状s的长度
    :return:
    '''
    x_train_lst = df["规范化后症状"].values
    y_train_lst = df["规范方剂"].values

    # 将x_train_lst中的顿号替换为空格
    x_train_lst_space = []
    symptoms_list = [] # 用于统计症状数量
    for symptoms in x_train_lst:
        x_train_lst_space.append(str(symptoms).replace("、", " "))
        symptoms_list.extend(symptoms.split("、"))
    # print(x_train_lst_space)
    print(len(x_train_lst_space))
    x_train_lst = np.array(x_train_lst_space)
    print(x_train_lst)

    # dictionary_size = 1000  # 词典大小
    dictionary_size = len(set(symptoms_list)) # 症状总数量，即词典大小
    tokenizer = Tokenizer(num_words=dictionary_size, lower=False)
    tokenizer.fit_on_texts(x_train_lst_space)
    # 为所有症状分配索引值，完成分词工作
    x_train_tokenized_lst = tokenizer.texts_to_sequences(x_train_lst_space)

    x_train = pad_sequences(x_train_tokenized_lst, maxlen=max_symptoms_length)
    #最后预测层输出维度
    output_dis = len(list(set(y_train_lst)))

    return x_train, y_train_lst, output_dis, dictionary_size

if __name__ == '__main__':
    # 读取数据集
    main_path = 'D:/mypython/GNN_yian'
    df = pd.read_csv(main_path + '/data/医案数据集utf8-3_症状规范化后-top-2.csv') #方剂规范化后，共有247个方剂;症状经过规范化，共459个症状
    df = filter_yian(df, freq_num=40) # 去除频次<=3的方的医案，剩余医案数为2901
    # dictionary_size = 1000 # 这应该是症状的总数量，也就是字典的大小
    max_symptoms_length = 20 #最多取多少个症状，也就是句子的长度
    x_train, y_train, output_dimension, dictionary_size = get_train_data(df, max_symptoms_length)

    # 构建模型
    embedding_vector_length = 256 # 一个症状嵌入的维度，也就是词嵌入向量的维度
    embdding = Embedding(dictionary_size, embedding_vector_length, input_length=max_symptoms_length)
    model = Sequential()
    model.add(embdding)
    model.add(SimpleRNN(100))
    model.add(Dense(10,  activation="relu")) #加入全连接层
    model.add(Dense(output_dimension,activation="softmax")) #加入分类输出层
    model.build(input_shape=(None, max_symptoms_length))
    model.compile(
        optimizer=tf.keras.optimizers.Adam(),
        # loss='binary_crossentropy',
        loss="sparse_categorical_crossentropy",
        metrics=['accuracy']
    )

    print(model.summary())

    # 训练
    history = model.fit(x_train, y_train,
                       validation_split=0.3,
                        epochs=40,
                        batch_size=64
                        )
    print(history)

2.模型的输入维度

训练时候输入的维度、中间层Dense的维度很容易出错，输入维度要正确，才能正确送入计算；最后一层的维度需要与标签的种类数一致，才能正确分类输出。

2.1输入维度

嵌入的维度：
NLP的任务，需要把字词嵌入表示为向量，因此训练之前，需要定义好嵌入（Embedding）的各个维度。

    embedding_vector_length = 128 # 一个症状嵌入的维度，也就是词嵌入向量的维度
    embdding = Embedding(dictionary_size, embedding_vector_length, input_length=max_symptoms_length)

dictionary_size指的是词表的大小，在这里是当前症状集合的大小，也就是说当前要把词在词表的范围内进行表示
embedding_vector_length是每个词向量的维度，在这里是每一个症状向量的维度，也就是说我们用多长的一个向量来表示一个症状，这里用的是128的长度的向量来表示
input_length=max_symptoms_length是一个样本的输入序列的长度，这里的输入序列指的就是一个医案中包含的多个症状，在其他文本任务重，指的可能是句子的长度；也就是一个样本中的那个x的长度。

model.add(SimpleRNN(100))

SimpleRNN层的维度是100，也就是100个神经元

2.2输出维度

    model.add(Dense(64,  activation="relu")) #加入全连接层
    model.add(Dense(output_dimension,activation="softmax")) #加入分类输出层

第一个全连接层的维度为64，即64个神经元
这里要注意的是，输出的维度（model.add(Dense(output_dimension,activation="softmax")) #加入分类输出层 中的output_dimension参数）必须与y的标签类型数量一致，不然是无法完成输出类型与你的y的分类类型匹配。

model.build(input_shape=(None, max_symptoms_length))

这里需要build模型，不然在新版的Keras中会报错，而且输入维度要写正确。
input_shape=(None, max_symptoms_length)的None参数其实是整个数据集的大小，在这里是医案的数量，这个地方可以是None，后面模型会根据数据自动算出来，
第二个参数max_symptoms_length指的是数据集的第二个维度，也就是x的长度，即一个样本的x长度，这里就是一个医案中的症状数量。
这里不需要再给第三个参数，如果给了：

model.build(input_shape=(None, max_symptoms_length,128))#这里加入128参数则报错

就会报下面的错误：

2023-11-25 10:36:46.789518: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX AVX2
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
Traceback (most recent call last):
  File "D:\mypython\GNN_yian\TCMPR\TCM_Formula_SimpleRNN.py", line 106, in <module>
    model.build(input_shape=(None, max_symptoms_length,128))
  File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\sequential.py", line 349, in build
    super(Sequential, self).build(input_shape)
  File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\training.py", line 449, in build
    self.call(x, **kwargs)
  File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\sequential.py", line 388, in call
    outputs = layer(inputs, **kwargs)
  File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\layers\rnn\base_rnn.py", line 515, in __call__
    return super(RNN, self).__call__(inputs, **kwargs)
  File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\utils\traceback_utils.py", line 67, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\input_spec.py", line 214, in assert_input_compatibility
    raise ValueError(f'Input {input_index} of layer "{layer_name}" '
ValueError: Input 0 of layer "simple_rnn" is incompatible with the layer: expected ndim=3, found ndim=4. Full shape received: (None, 20, 128, 128)

错误提示：期待的维度是3，却给到了4
于是去掉128就行，即：

model.build(input_shape=(None, max_symptoms_length))

下面是整个model输出的维度信息：

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 module_wrapper (ModuleWrapp  (None, 20, 128)          44544     
 er)                                                             
                                                                 
 simple_rnn (SimpleRNN)      (None, 100)               22900     
                                                                 
 dense (Dense)               (None, 64)                6464      
                                                                 
 dense_1 (Dense)             (None, 10)                650  

参考文献：

[1]深度学习的特征与标签的维度（keras)

[2]【机器学习】 - 关于Keras的深入理解

[3]NLP学习笔记（四）：关于keras的Input层与embedding层全解析

[4]用tensorflow.keras写神经网络处理nlp遇到的问题

3.标签为字符串的问题

上面代码在运行后，报错为：

2023-11-24 15:42:20.697620: W tensorflow/core/framework/op_kernel.cc:1722] OP_REQUIRES failed at cast_op.cc:121 : UNIMPLEMENTED: Cast string to float is not supported
Traceback (most recent call last):
  File "D:\mypython\GNN_yian\TCMPR\TCM_Formula_SimpleRNN.py", line 107, in <module>
    history = model.fit(x_train, y_train,
  File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\utils\traceback_utils.py", line 67, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\tensorflow\python\eager\execute.py", line 54, in quick_execute
    tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
tensorflow.python.framework.errors_impl.UnimplementedError: Graph execution error:

Detected at node 'Cast_1' defined at (most recent call last):
    File "D:\mypython\GNN_yian\TCMPR\TCM_Formula_SimpleRNN.py", line 107, in <module>
      history = model.fit(x_train, y_train,
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\utils\traceback_utils.py", line 64, in error_handler
      return fn(*args, **kwargs)
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\training.py", line 1409, in fit
      tmp_logs = self.train_function(iterator)
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\training.py", line 1051, in train_function
      return step_function(self, iterator)
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\training.py", line 1040, in step_function
      outputs = model.distribute_strategy.run(run_step, args=(data,))
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\training.py", line 1030, in run_step
      outputs = model.train_step(data)
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\training.py", line 894, in train_step
      return self.compute_metrics(x, y, y_pred, sample_weight)
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\training.py", line 987, in compute_metrics
      self.compiled_metrics.update_state(y, y_pred, sample_weight)
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\engine\compile_utils.py", line 501, in update_state
      metric_obj.update_state(y_t, y_p, sample_weight=mask)
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\utils\metrics_utils.py", line 70, in decorated
      update_op = update_state_fn(*args, **kwargs)
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\metrics\base_metric.py", line 140, in update_state_fn
      return ag_update_state(*args, **kwargs)
    File "D:\anaconda3\envs\pytorch_cpu\lib\site-packages\keras\metrics\base_metric.py", line 637, in update_state
      y_true = tf.cast(y_true, self._dtype)
Node: 'Cast_1'
Cast string to float is not supported
     [[{{node Cast_1}}]] [Op:__inference_train_function_1551]

Process finished with exit code 1

"UnimplementedError: Cast string to float is not supported" 错误是由于在 TensorFlow 中尝试将字符串转换为浮点数时出错导致的。这可能是因为您的数据集中包含不是浮点数的数据类型，例如字符串或布尔值。
代码中y_train是“方剂名称”，类型是字符串，故报错。
将方剂名称转换为数字，映射到对应的数字上。
增加一个函数：

def y_train_mapping(y_train):
    '''
    将字符串的y_train，映射到其对应的index上。（如果不映射，在TensorFlow中会报错）
    :param y_train:
    :return:
    '''
    y_train_no_re = list(set(y_train)) # 去重后的方剂列表
    mapping_dict = {} # key是方剂名，value是其index
    for i, y in enumerate(y_train_no_re):
        mapping_dict[y] = i
    y_train_ = [mapping_dict.get(y) for y in y_train]
    return np.array(y_train_), mapping_dict

参考文献：

[1]详解TensorFlow报”UnimplementedError: Cast string to float is not supported “的原因以及解决办法

[2]TensorFlow无法将字符串转换为浮点错误？