Sci-Kit bug Learn SGD Algo - “Array contains NaN or infinity”

Question

I get the error "Array contains NaN or infinity." I checked my data both for the train / test for the absence of values, and for the absence of errors.

Perhaps I have a misinterpretation of what “array containing NaN or infinity” means.

import numpy as np from sklearn import linear_model from numpy import genfromtxt, savetxt def main(): #create the training & test sets, skipping the header row with [1:] dataset = genfromtxt(open('C:\\Users\\Owner\\training.csv','r'), delimiter=',')[0:50] target = [x[0] for x in dataset] train = [x[1:50] for x in dataset] test = genfromtxt(open('C:\\Users\\Owner\\test.csv','r'), delimiter=',')[0:50] #create and train the SGD sgd = linear_model.SGDClassifier() sgd.fit(train, target) predictions = [x[1] for x in sgd.predict(test)] savetxt('C:\\Users\\Owner\\Desktop\\preds.csv', predictions, delimiter=',', fmt='%f') if __name__=="__main__": main() 

I thought that the data type could call an algorithm for the loop (they are floating points).

I know that SGD can handle floating points, so I'm not sure if this setting requires a data type to be declared.

For example, one of the following:

 >>> dt = np.dtype('i4') # 32-bit signed integer >>> dt = np.dtype('f8') # 64-bit floating-point number >>> dt = np.dtype('c16') # 128-bit complex floating-point number >>> dt = np.dtype('a25') # 25-character string 

The following is a complete error message:

 --------------------------------------------------------------------------- ValueError Traceback (most recent call last) <ipython-input-62-af5537e7802b> in <module>() 19 20 if __name__=="__main__": ---> 21 main() <ipython-input-62-af5537e7802b> in main() 13 #create and train the SGD 14 sgd = linear_model.SGDClassifier() ---> 15 sgd.fit(train, target) 16 predictions = [x[1] for x in sgd.predict(test)] 17 C:\Anaconda\lib\site-packages\sklearn\linear_model\stochastic_gradient.pyc in fi t(self, X, y, coef_init, intercept_init, class_weight, sample_weight) 518 coef_init=coef_init, intercept_init=intercept_i nit, 519 class_weight=class_weight, --> 520 sample_weight=sample_weight) 521 522 C:\Anaconda\lib\site-packages\sklearn\linear_model\stochastic_gradient.pyc in _f it(self, X, y, alpha, C, loss, learning_rate, coef_init, intercept_init, class_w eight, sample_weight) 397 self.class_weight = class_weight 398 --> 399 X = atleast2d_or_csr(X, dtype=np.float64, order="C") 400 n_samples, n_features = X.shape 401 C:\Anaconda\lib\site-packages\sklearn\utils\validation.pyc in atleast2d_or_csr(X , dtype, order, copy) 114 """ 115 return _atleast2d_or_sparse(X, dtype, order, copy, sparse.csr_matrix , --> 116 "tocsr") 117 118 C:\Anaconda\lib\site-packages\sklearn\utils\validation.pyc in _atleast2d_or_spar se(X, dtype, order, copy, sparse_class, convmethod) 94 _assert_all_finite(X.data) 95 else: ---> 96 X = array2d(X, dtype=dtype, order=order, copy=copy) 97 _assert_all_finite(X) 98 return X C:\Anaconda\lib\site-packages\sklearn\utils\validation.pyc in array2d(X, dtype, order, copy) 79 'is required. Use X.toarray() to convert to dens e.') 80 X_2d = np.asarray(np.atleast_2d(X), dtype=dtype, order=order) ---> 81 _assert_all_finite(X_2d) 82 if X is X_2d and copy: 83 X_2d = safe_copy(X_2d) C:\Anaconda\lib\site-packages\sklearn\utils\validation.pyc in _assert_all_finite (X) 16 if (X.dtype.char in np.typecodes['AllFloat'] and not np.isfinite(Xs um()) 17 and not np.isfinite(X).all()): ---> 18 raise ValueError("Array contains NaN or infinity.") 19 20 ValueError: Array contains NaN or infinity. 

Any thoughts would be appreciated.

Answer 1

As reporting an error, there is either np.nan in your data, or np.inf or -np.inf . Since you are reading from a text file and you are saying that your data does not contain missing values, this is probably caused by column headers or some other value in the file that cannot be automatically interpreted.

The documentation for genfromtxt shows that by default, dtype for float read arrays, which means that all the values ​​you read should pass the equivalent of float(x) .

If you are not sure if this is causing an error, you can remove non-finite numbers from the numpy array as follows:

 dataset[ ~np.isfinite(dataset) ] = 0 # Set non-finite (nan, inf, -inf) to zero 

If this fixes the error, you can be sure that you have incorrect values ​​in your variable. To find where you can use the following:

 np.where(~np.isfinite(dataset)) 

This will return a list of indexes that contain invalid values, e.g.

 >>> import numpy as np >>> dataset = np.array([[0,1,1],[np.nan,0,0],[1,2,np.inf]]) >>> dataset array([[ 0., 1., 1.], [ nan, 0., 0.], [ 1., 2., inf]]) >>> np.where(~np.isfinite(dataset)) (array([1, 2]), array([0, 2]))