6
votes

I have been trying this:

  1. Create X features and y dependent from a dataset
  2. Split the dataset
  3. Normalise the data
  4. Train using SVR from Scikit-learn

Here is the code using a pandas dataframe filled with random values

import pandas as pd
import numpy as np
df = pd.DataFrame(np.random.rand(20,5), columns=["A","B","C","D", "E"])
a = list(df.columns.values)
a.remove("A")

X = df[a]
y = df["A"]

X_train = X.iloc[0: floor(2 * len(X) /3)]
X_test = X.iloc[floor(2 * len(X) /3):]
y_train = y.iloc[0: floor(2 * len(y) /3)]
y_test = y.iloc[floor(2 * len(y) /3):]

# normalise

from sklearn import preprocessing

X_trainS = preprocessing.scale(X_train)
X_trainN = pd.DataFrame(X_trainS, columns=a)

X_testS = preprocessing.scale(X_test)
X_testN = pd.DataFrame(X_testS, columns=a)

y_trainS = preprocessing.scale(y_train)
y_trainN = pd.DataFrame(y_trainS)

y_testS = preprocessing.scale(y_test)
y_testN = pd.DataFrame(y_testS)

import sklearn
from sklearn.svm import SVR

clf = SVR(kernel='rbf', C=1e3, gamma=0.1)

pred = clf.fit(X_trainN,y_trainN).predict(X_testN)

gives this error:

C:\Anaconda3\lib\site-packages\pandas\core\index.py:542: FutureWarning: slice indexers when using iloc should be integers and not floating point "and not floating point",FutureWarning) --------------------------------------------------------------------------- ValueError Traceback (most recent call last) in () 34 clf = SVR(kernel='rbf', C=1e3, gamma=0.1) 35 ---> 36 pred = clf.fit(X_trainN,y_trainN).predict(X_testN) 37

C:\Anaconda3\lib\site-packages\sklearn\svm\base.py in fit(self, X, y, sample_weight) 174 175 seed = rnd.randint(np.iinfo('i').max) --> 176 fit(X, y, sample_weight, solver_type, kernel, random_seed=seed) 177 # see comment on the other call to np.iinfo in this file 178

C:\Anaconda3\lib\site-packages\sklearn\svm\base.py in _dense_fit(self, X, y, sample_weight, solver_type, kernel, random_seed) 229 cache_size=self.cache_size, coef0=self.coef0, 230 gamma=self._gamma, epsilon=self.epsilon, --> 231 max_iter=self.max_iter, random_seed=random_seed) 232 233 self._warn_from_fit_status()

C:\Anaconda3\lib\site-packages\sklearn\svm\libsvm.pyd in sklearn.svm.libsvm.fit (sklearn\svm\libsvm.c:1864)()

ValueError: Buffer has wrong number of dimensions (expected 1, got 2)

I am not sure why. Can anyone explain? I think it has something to with converting back to dataframes after preprocessing.

1
The error is in your y_trainN, it's producing an incorrect array shape the following works: pred = clf.fit(X_trainN,y_trainN.squeeze().values).predict(X_testN), if you look at what is outputted when you do y_trainN.values you get a nested array when what you want is just an array even though you have just a single column in your df, what you should do is pass a single column: pred = clf.fit(X_trainN,y_trainN[0]).predict(X_testN)EdChum
Also you can just do a = list(df) rather than a = list(df.columns.values) if you want a list of the columns see related post.EdChum
thanks that is really helpful. Think you have answered a lot of my questions today!azuric
I got a similar error. The problem was that I was using 1 hot vectors as y instead of class numbers.Souradeep Nanda

1 Answers

4
votes

The error here is in the df you pass as your labels: y_trainN

if you compare against the sample docs version and your code:

In [40]:

n_samples, n_features = 10, 5
np.random.seed(0)
y = np.random.randn(n_samples)
print(y)
y_trainN.values
[ 1.76405235  0.40015721  0.97873798  2.2408932   1.86755799 -0.97727788
  0.95008842 -0.15135721 -0.10321885  0.4105985 ]
Out[40]:
array([[-0.06680594],
       [ 0.23535043],
       [-1.49265082],
       [ 1.22537862],
       [-0.46499134],
       [-0.23744759],
       [ 1.40520679],
       [ 0.95882677],
       [ 1.66996413],
       [-0.37515955],
       [-0.75826444],
       [-1.45945337],
       [-0.63995369]])

So you can either call squeeze to produce a series or select the only column in the df in order for there to be no errors:

pred = clf.fit(X_trainN,y_trainN[0]).predict(X_testN)

or

pred = clf.fit(X_trainN,y_trainN.squeeze()).predict(X_testN)

so we could argue that for a df with only a single column it should return something that can then be coerced into a numpy array or that numpy is not calling the array attribute correctly but really you should pass a series or select the column from a df as the params