Numpy hstack - "ValueError: all the input arrays must have same number of dimensions" - but they do

Question

I am trying to join two numpy arrays. In one I have a set of columns/features after running TF-IDF on a single column of text. In the other I have one column/feature which is an integer. So I read in a column of train and test data, run TF-IDF on this, and then I want to add another integer column because I think this will help my classifier learn more accurately how it should behave.

Unfortunately, I am getting the error in the title when I try and run hstack to add this single column to my other numpy array.

Here is my code :

  #reading in test/train data for TF-IDF
  traindata = list(np.array(p.read_csv('FinalCSVFin.csv', delimiter=";"))[:,2])
  testdata = list(np.array(p.read_csv('FinalTestCSVFin.csv', delimiter=";"))[:,2])

  #reading in labels for training
  y = np.array(p.read_csv('FinalCSVFin.csv', delimiter=";"))[:,-2]

  #reading in single integer column to join
  AlexaTrainData = p.read_csv('FinalCSVFin.csv', delimiter=";")[["alexarank"]]
  AlexaTestData = p.read_csv('FinalTestCSVFin.csv', delimiter=";")[["alexarank"]]
  AllAlexaAndGoogleInfo = AlexaTestData.append(AlexaTrainData)

  tfv = TfidfVectorizer(min_df=3,  max_features=None, strip_accents='unicode',  
        analyzer='word',token_pattern=r'\w{1,}',ngram_range=(1, 2), use_idf=1,smooth_idf=1,sublinear_tf=1) #tf-idf object
  rd = lm.LogisticRegression(penalty='l2', dual=True, tol=0.0001, 
                             C=1, fit_intercept=True, intercept_scaling=1.0, 
                             class_weight=None, random_state=None) #Classifier
  X_all = traindata + testdata #adding test and train data to put into tf-idf
  lentrain = len(traindata) #find length of train data
  tfv.fit(X_all) #fit tf-idf on all our text
  X_all = tfv.transform(X_all) #transform it
  X = X_all[:lentrain] #reduce to size of training set
  AllAlexaAndGoogleInfo = AllAlexaAndGoogleInfo[:lentrain] #reduce to size of training set
  X_test = X_all[lentrain:] #reduce to size of training set

  #printing debug info, output below : 
  print "X.shape => " + str(X.shape)
  print "AllAlexaAndGoogleInfo.shape => " + str(AllAlexaAndGoogleInfo.shape)
  print "X_all.shape => " + str(X_all.shape)

  #line we get error on
  X = np.hstack((X, AllAlexaAndGoogleInfo))

Below is the output and error message :

X.shape => (7395, 238377)
AllAlexaAndGoogleInfo.shape => (7395, 1)
X_all.shape => (10566, 238377)



---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-12-2b310887b5e4> in <module>()
     31 print "X_all.shape => " + str(X_all.shape)
     32 #X = np.column_stack((X, AllAlexaAndGoogleInfo))
---> 33 X = np.hstack((X, AllAlexaAndGoogleInfo))
     34 sc = preprocessing.StandardScaler().fit(X)
     35 X = sc.transform(X)

C:\Users\Simon\Anaconda\lib\site-packages\numpy\core\shape_base.pyc in hstack(tup)
    271     # As a special case, dimension 0 of 1-dimensional arrays is "horizontal"
    272     if arrs[0].ndim == 1:
--> 273         return _nx.concatenate(arrs, 0)
    274     else:
    275         return _nx.concatenate(arrs, 1)

ValueError: all the input arrays must have same number of dimensions

What is causing my problem here? How can I fix this? As far as I can see I should be able to join these columns? What have I misunderstood?

Thank you.

Edit :

Using the method in the answer below gets the following error :

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-16-640ef6dd335d> in <module>()
---> 36 X = np.column_stack((X, AllAlexaAndGoogleInfo))
     37 sc = preprocessing.StandardScaler().fit(X)
     38 X = sc.transform(X)

C:\Users\Simon\Anaconda\lib\site-packages\numpy\lib\shape_base.pyc in column_stack(tup)
    294             arr = array(arr,copy=False,subok=True,ndmin=2).T
    295         arrays.append(arr)
--> 296     return _nx.concatenate(arrays,1)
    297 
    298 def dstack(tup):

ValueError: all the input array dimensions except for the concatenation axis must match exactly

Interestingly, I tried to print the dtype of X and this worked fine :

X.dtype => float64

However, trying to print the dtype of AllAlexaAndGoogleInfo like so :

print "AllAlexaAndGoogleInfo.dtype => " + str(AllAlexaAndGoogleInfo.dtype) 

produces :

'DataFrame' object has no attribute 'dtype'

Answer 1

As X is a sparse array, instead of numpy.hstack, use scipy.sparse.hstack to join the arrays. In my opinion the error message is kind of misleading here.

This minimal example illustrates the situation:

import numpy as np
from scipy import sparse

X = sparse.rand(10, 10000)
xt = np.random.random((10, 1))
print 'X shape:', X.shape
print 'xt shape:', xt.shape
print 'Stacked shape:', np.hstack((X,xt)).shape
#print 'Stacked shape:', sparse.hstack((X,xt)).shape #This works

Based on the following output

X shape: (10, 10000)
xt shape: (10, 1)

one may expect that the hstack in the following line will work, but the fact is that it throws this error:

ValueError: all the input arrays must have same number of dimensions

So, use scipy.sparse.hstack when you have a sparse array to stack.

---

In fact I have answered this as a comment in your another questions, and you mentioned that another error message pops up:

TypeError: no supported conversion for types: (dtype('float64'), dtype('O'))

First of all, AllAlexaAndGoogleInfo does not have a dtype as it is a DataFrame. To get it's underlying numpy array, simply use AllAlexaAndGoogleInfo.values. Check its dtype. Based on the error message, it has a dtype of object, which means that it might contain non-numerical elements like strings.

This is a minimal example that reproduces this situation:

X = sparse.rand(100, 10000)
xt = np.random.random((100, 1))
xt = xt.astype('object') # Comment this to fix the error
print 'X:', X.shape, X.dtype
print 'xt:', xt.shape, xt.dtype
print 'Stacked shape:', sparse.hstack((X,xt)).shape

The error message:

TypeError: no supported conversion for types: (dtype('float64'), dtype('O'))

So, check if there is any non-numerical values in AllAlexaAndGoogleInfo and repair them, before doing the stacking.

Answer 2

Use .column_stack. Like so:

X = np.column_stack((X, AllAlexaAndGoogleInfo))

From the docs:

Take a sequence of 1-D arrays and stack them as columns to make a single 2-D array. 2-D arrays are stacked as-is, just like with hstack.