问题描述
### using trasnformers
from sklearn.compose import ColumnTransformer
from sklearn.feature_extraction.text import CountVectorizer,TfidfVectorizer
from sklearn.preprocessing import OneHotEncoder
from sklearn.preprocessing import MinMaxScaler
column_trans = ColumnTransformer(
[
('CompanyName_bow',TfidfVectorizer(),'CompanyName'),('state_category',OneHotEncoder(),['state']),('Termination_Reason_Desc_bow','Termination_Reason_Desc'),('TermType_category',['TermType'])
],remainder=MinMaxScaler()
)
X = column_trans.fit_transform(X.head(100))
from sklearn.preprocessing import LabelEncoder
y = LabelEncoder().fit_transform(y.head(100))
from sklearn.model_selection import train_test_split
X_train,X_test,y_train,y_test = train_test_split(X,y,test_size=0.2,random_state=5)
X_train.shape #(80,92)
X_test.shape #(20,92)
y_train.shape #(80,)
X_train.todense()
matrix([[0.,0.,...,0.26921709,1.,0. ],[0.,1. ],0.46148896,0. ]])
type(X_train)
--> scipy.sparse.csr.csr_matrix
print(y_train)
array([0,1,0])
type(y_train)
numpy.ndarray
# use autokeras to find a model for the sonar dataset
from numpy import asarray
from pandas import read_csv
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder
from autokeras import StructuredDataClassifier
print(X_train.shape,X_test.shape,y_train.shape,y_test.shape)
# define the search
search = StructuredDataClassifier(max_trials=15)
# perform the search
search.fit(x=(X_train),y=y_train,verbose=0)
# evaluate the model
loss,acc = search.evaluate(X_test,y_test,verbose=0)
print('Accuracy: %.3f' % acc)
(80,92) (20,92) (80,) (20,)
INFO:tensorflow:Reloading Oracle from existing project .\structured_data_classifier\oracle.json
INFO:tensorflow:Reloading Tuner from .\structured_data_classifier\tuner0.json
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-106-94708e5d279d> in <module>
10 search = StructuredDataClassifier(max_trials=15)
11 # perform the search
---> 12 search.fit(x=(X_train),verbose=0)
13 # evaluate the model
14 loss,verbose=0)
~\anaconda3\lib\site-packages\autokeras\tasks\structured_data.py in fit(self,x,epochs,callbacks,validation_split,validation_data,**kwargs)
313 [keras.Model.fit](https://www.tensorflow.org/api_docs/python/tf/keras/Model#fit).
314 """
--> 315 super().fit(
316 x=x,317 y=y,~\anaconda3\lib\site-packages\autokeras\tasks\structured_data.py in fit(self,**kwargs)
132 self.check_in_fit(x)
133
--> 134 super().fit(
135 x=x,136 y=y,~\anaconda3\lib\site-packages\autokeras\auto_model.py in fit(self,batch_size,**kwargs)
259 validation_split = 0
260
--> 261 dataset,validation_data = self._convert_to_dataset(
262 x=x,y=y,validation_data=validation_data,batch_size=batch_size
263 )
~\anaconda3\lib\site-packages\autokeras\auto_model.py in _convert_to_dataset(self,batch_size)
373 x = dataset.map(lambda x,y: x)
374 y = dataset.map(lambda x,y: y)
--> 375 x = self._adapt(x,self.inputs,batch_size)
376 y = self._adapt(y,self._heads,batch_size)
377 dataset = tf.data.Dataset.zip((x,y))
~\anaconda3\lib\site-packages\autokeras\auto_model.py in _adapt(self,dataset,hms,batch_size)
287 adapted = []
288 for source,hm in zip(sources,hms):
--> 289 source = hm.get_adapter().adapt(source,batch_size)
290 adapted.append(source)
291 if len(adapted) == 1:
~\anaconda3\lib\site-packages\autokeras\engine\adapter.py in adapt(self,batch_size)
65 tf.data.Dataset. The converted dataset.
66 """
---> 67 self.check(dataset)
68 dataset = self.convert_to_dataset(dataset,batch_size)
69 return dataset
~\anaconda3\lib\site-packages\autokeras\adapters\input_adapters.py in check(self,x)
63 def check(self,x):
64 if not isinstance(x,(pd.DataFrame,np.ndarray,tf.data.Dataset)):
---> 65 raise TypeError(
66 "Unsupported type {type} for "
67 "{name}.".format(type=type(x),name=self.__class__.__name__)
TypeError: Unsupported type <class 'scipy.sparse.csr.csr_matrix'> for StructuredDataAdapter.
解决方法
正如您在与此线程并行打开的 Github issue 中所注意到的,AutoKeras(当前)不支持稀疏矩阵,建议将它们转换为密集的 Numpy 数组。实际上,从 AutoKeras StructuredDataClassifier
的 documentation 来看,相应 x
方法中的训练数据 .fit
预计为:
字符串、numpy.ndarray、pandas.DataFrame 或 tensorflow.Dataset
而不是 SciPy 稀疏矩阵。
鉴于此处您的 X_train
非常小:
X_train.shape
# (80,92)
您完全没有理由使用稀疏矩阵。尽管在这里您似乎试图将 X_train
转换为密集的,但您没有重新分配它,结果是它仍然是一个稀疏的;来自您自己的上述代码:
X_train.todense()
# ...
type(X_train)
# scipy.sparse.csr.csr_matrix
你需要做的只是将它重新分配给一个密集数组:
from scipy.sparse import csr_matrix
X_train = X_train.toarray()
这是一个使用虚拟数据的简短演示:
import numpy as np
from scipy.sparse import csr_matrix
X_train = csr_matrix((3,4),dtype=np.float)
type(X_train)
# scipy.sparse.csr.csr_matrix
# this will not work:
X_train.todense()
type(X_train)
# scipy.sparse.csr.csr_matrix # still sparse
# this will work:
X_train = X_train.toarray()
type(X_train)
# numpy.ndarray
您应该对 X_test
数据执行类似的过程(您的 y_train
和 y_test
似乎已经是密集的 Numpy 数组)。
可能的原因(由于上面粘贴的代码可读性低)可能是使用不同的数据集和保存的模型。我建议您在 overwrite=True
构造代码块中添加 BayesianOptimization
。重新安装 TensorFlow 也可能有所帮助。