NumPy数组转换报错：形状不均匀及Keras兼容问题求助

问题背景 #

运行聊天机器人训练代码时，第22行training = np.array(training)触发以下ValueError：

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 2 dimensions. The detected shape was (810, 2) + inhomogeneous part.

该代码一年前可正常运行，更新NumPy后与新版Keras出现兼容问题，需修复错误或获取NumPy用法指导。

相关代码及报错回溯 #

核心报错代码片段 #

# initializing training data
training = []
output_empty = [0] * len(classes)
for doc in documents:
    # initializing bag of words
    bag = []
    # list of tokenized words for the pattern
    pattern_words = doc[0]
    # lemmatize each word - create base word, in attempt to represent related words
    pattern_words = [lemmatizer.lemmatize(word.lower()) for word in pattern_words]
    # create our bag of words array with 1, if word match found in current pattern
    for w in words:
        bag.append(1) if w in pattern_words else bag.append(0)
    # output is a '0' for each tag and '1' for current tag (for each pattern)
    output_row = list(output_empty)
    output_row[classes.index(doc[1])] = 1
    training.append([bag, output_row])
# shuffle our features and turn into np.array
random.shuffle(training)
training = np.array(training)  # 报错行
# create train and test lists. X - patterns, Y - intents
train_x = list(training[:,0])
train_y = list(training[:,1])

完整报错回溯 #

ValueError                                Traceback (most recent call last)
<ipython-input-8-1a891a7a8859> in <cell line: 22>()
     20 # shuffle our features and turn into np.array
     21 random.shuffle(training)
---&gt; 22 training = np.array(training)
     23 # create train and test lists. X - patterns, Y - intents
     24 train_x = list(training[:,0])

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 2 dimensions. The detected shape was (810, 2) + inhomogeneous part.

修复方案 #

方案一：拆分特征与标签生成流程（推荐） #

新版NumPy对数组形状一致性检查更严格，不再允许隐式转换不均匀嵌套列表。直接分开处理特征（train_x）和标签（train_y），避免创建形状不均匀的中间数组：

修改后的代码片段：

# 初始化训练数据，分开存储特征和标签
train_x = []
train_y = []
output_empty = [0] * len(classes)
for doc in documents:
    # 初始化词袋
    bag = []
    pattern_words = doc[0]
    pattern_words = [lemmatizer.lemmatize(word.lower()) for word in pattern_words]
    # 创建词袋数组
    for w in words:
        bag.append(1) if w in pattern_words else bag.append(0)
    # 创建输出标签行
    output_row = list(output_empty)
    output_row[classes.index(doc[1])] = 1
    # 分别添加到特征和标签列表
    train_x.append(bag)
    train_y.append(output_row)
# 同步打乱特征和标签
combined = list(zip(train_x, train_y))
random.shuffle(combined)
train_x, train_y = zip(*combined)
# 转换为规整的NumPy数组
train_x = np.array(train_x)
train_y = np.array(train_y)

此方式完全适配新版NumPy，代码逻辑更清晰，避免不必要的中间转换。

方案二：强制转换为对象类型数组（临时兼容） #

若需保留原有合并结构，可通过指定dtype=object强制NumPy接受不均匀数组，但后续仍需额外转换才能用于模型训练：

training = np.array(training, dtype=object)
train_x = np.array(list(training[:,0]))
train_y = np.array(list(training[:,1]))

该方案仅作为临时兼容手段，性能和代码可读性均不如方案一。

问题根源说明 #

旧版NumPy会自动将不均匀嵌套列表转换为包含对象的数组，而新版NumPy强化了形状检查，抛出明确错误以避免隐式转换带来的性能损耗和潜在bug。方案一的显式拆分处理是符合现代NumPy最佳实践的解决方式。

内容的提问来源于stack exchange，提问作者Vikash Kumar Mehta