学习使用bilstm

1. TypeError: Tensors in list passed to 'values' of 'ConcatV2' Op have types [bool, float32] that don't all match.

出现地方：

bilstm_crf_model = Embedding(input_dim=n_words + 1, output_dim=20, input_length=max_len, mask_zero=True)(input) # 20-dim embedding

解决：删掉mask_zero=True

bilstm_crf_model = Embedding(input_dim=n_words + 1, output_dim=20, input_length=max_len)(input) # 20-dim embedding

2.TensorFlow与Keras，CUDA版本问题

2.1 AttributeError: module 'keras.backend' has no attribute 'slice'

2.2 attributeerror: 'node' object has no attribute 'output_masks'

2.3 typeerror: __init__() got an unexpected keyword argument 'ragged'

2.4 keras_contrib attributeerror: module 'tensorflow' has no attribute 'get_default_graph'

2.5 importerror: cannot import name 'saving'

2.6 futurewarning: from version 0.24, get_params will raise an attributeerror if a parameter cannot be retrieved as an instance attribute. previously it would return none. futurewarning)

TensorFlow与Keras，CUDA，CUDNN，Python相应版本相对应很重要

我这边目前是：CUDA Version 10.0.130，CUDNN：7.5 ，Python 3.6.10，GCC 7.3.0

我安装的： TensorFlow：2.0.0，Keras：2.3.1

查看CUDA的方法：

第一种.：nvidia-smi

第二种：

cat /usr/local/cuda/version.txt

查看CUDNN的方法：

cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

查看python，GCC版本：

python 

CUDA，CUDNN与TensorFlow版本对应：https://tensorflow.google.cn/install/source

TensorFlow与Keras版本对应：https://docs.floydhub.com/guides/environments/

3.KeyError: 'accuracy'

出现地方：

model.compile(optimizer="rmsprop", loss=crf.loss_function, metrics=["accuracy"]) history = model.fit(X_train, np.array(y_train), batch_size=32, epochs=5, validation_split=0.1, verbose=1) hist = pd.DataFrame(history.history) plt.figure(figsize=(12,12)) plt.plot(hist["acc"]) plt.plot(hist["val_acc"]) plt.show()

解决：列名要注意

 

plt.plot(hist["accuracy"]) plt.plot(hist["val_accuracy"])

 4.typeerror: load_model() missing 1 required positional argument: 'filepath'

出现地方： model=load_model(filepath="../result/bi-lstm.h5")

解决：

filepath = "result/bi-lstm.h5" model = load_model(filepath)

5.attributeerror: 'float' object has no attribute 'lower'

出现原因：遇到了数据为NaN的情况

解决：用空格去填充缺失值

data=data.fillna('')