为什么Python中的LIBSVM要求我提供浮动值？

Question

I'm implementing an SVM classification problem using LIBSVM in python. I have a numpy array of consisting of 1.0 and -1.0 called train_labels and the corresponding features in another numpy array called train_data. Since LIBSVM does not accept numpy arrays, I convert them to lists using the code below.

train_labels = train_labels.tolist()
train_data = train_data.tolist()

However, when I put them on svm_problem as:

prob = svm_problem(train_labels,train_data)

I'm getting then error

  File "C:\Anaconda\lib\site-packages\svm.py", line 109, in __init__
  for i, yi in enumerate(y): self.y[i] = yi
TypeError: a float is required

I've already tried converting them to float using train_labels = train_labels.astype(np.float) before converting to list but I'm still getting the same error.

Using tolist() method for converting numpy array to lists before putting them on LIBSVM commands is working when I've tried them in the console.

Does anyone know why I'm getting this error? And how can I solve it?

Answer 1

guys! After some more idling in my code, I've found what's wrong.

The line train_labels = train_labels.tolist() is converting my array to a list of lists with one element in the sublist. I think this is because train_labels is an nx 1 array to start with (it is read from a csv file).

So I need to reshape it first before converting to list. And it works.

train_labels = train_labels.reshape(N_train).tolist()

I hope it's ok to still ask this question here for those starting to use LIBSVM in the future.

Thanks!