尝试连接keras模型：值错误：未能将NumPy数组转换为张量（不支持的对象类型浮点）

Question

我试图在keras中连接两个并行模型，每个模型都有不同的输入。相关代码如下。

# model 1
model1_in = Input(shape=(train_x_1.shape[1], train_x_1.shape[2]))
model1_out = LSTM(50, activation='relu',return_sequences=False, name='layer_1')(model1_in)
model1 = Model(model1_in, model1_out)

# model 2
model2_in = Input(shape=(1))
model2_out = Dense(8, activation='relu', name='layer_2')(model2_in)
model2 = Model(model2_in, model2_out)

concatenated = concatenate(inputs=[model1.output, model2.output])
out = Dense(1, activation='relu', name='output_layer')(concatenated)
model = Model([model1_in, model2_in], out)
model.compile(loss='mean_absolute_error', optimizer='adam')


# fit network
history = model.fit([train_x_1,train_x_2], train_y, epochs=100, batch_size=72, validation_data=([test_x_1,test_x_2], test_y), verbose=2, shuffle=False)

我得到的错误是

将NumPy数组转换为张量失败（不支持的对象类型浮点数）。

并在模型上发生。装配线。

我在空转。训练值和测试值都是数组，我检查了所有训练输入的长度是否相同：

#train_x_1.shape[0]
15465
#train_y.shape[0]
15465
#train_x_2.shape[0]
15465
#test_x_1.shape[0]
1719
#test_x_2.shape[0]
1719
#test_y.shape[0]
1719
#test_x_1
array([[[0.6243922 ],
        [0.5463666 ],
        [0.7083546 ], ... etc ...

任何帮助都将不胜感激-提前感谢！

完整的错误跟踪如下所示：

回溯（最近的调用最后）：文件filepath.py，行220，在历史=model.fit（[train_x_1，train_x_2]，train_y，时代=100，batch_size=72，validation_data=（[test_x_1，test_x_2]，test_y），冗长=2，洗牌=假）文件/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training.py"，第728行，适合use_multiprocessing=use_multiprocessing）文件"/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2.py"，第224行，适合distribution_strategy=策略）文件"/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2.py"，第547行，_process_training_inputs=use_multiprocessing）文件“/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2.py”，第606行，_process_inputsuse_multiprocessing=use_multiprocessing）文件“/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/data_adapter.py”，第217行，在init x=_process_numpy_inputs（x）文件“/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/data_adapter.py”，第703行，在_process_numpy_inputs输入=nest.map_structure（_convert_non_tensor，输入）文件/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/util/nest.py，第535行，map_structure结构[0]，条目中x的[func（*x）]，文件/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/util/nest.py，第535行，在结构[0]中，[func（*x）用于条目中的x]，文件"/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/data_adapter.py"，第700行，_convert_non_tensor返回ops.convert（x）文件"/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/framework/ops.py"，第1184行，在convert_to_tensor返回convert_to_tensor_v2（值，dtype，preferred_dtype，名称）文件"/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/framework/ops.py"，第1242行，convert_to_tensor_v2as_ref=False）文件"/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/framework/ops.py"，第1296行，internal_convert_to_tensorret=conversion_func（值，dtype=dtype，名称=名称，as_ref=as_ref）文件"/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/framework/tensor_conversion_registry.py"，第52行，_default_conversion_function返回constant_op.constant（值，dtype，名称=名称）文件"/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/framework/constant_op.py"，第227行，常量allow_broadcast=True）文件"/库/框架/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_core/python/framework/constant_op.py"，第235行，_constant_implt=convert_to_eager_tensor（value， ctx， dtype）File"/库/框架/Python.框架/版本/3.6/lib/python3.6/site-包/tensorflow_core/python/框架/constant_op. py"，第96行，convert_to_eager_tensor返回ops。EagerTensor（value， ctx.device_name， dtype）ValueError：无法将NumPy数组转换为张量（不支持的对象类型浮点数）。

Answer 1

为了社区的利益，在回答部分指定解决方案（即使它存在于评论部分）。

x_2的值均为typefloat，而x_1的值为float32。

使用将x2修改为Float32x2.astype（'Float32'）已解决此问题。