Scikit-learn (SVC estimator) always give the same value for predictions

Question

I am doing a task on supervised learning. I have two set of data -training and test.

My training data-set is about 2000 records. My test data has 10 records.
When I run the following code, the predict function gives the same value as output.

I am not sure what I am doing wrong... I tried changing the value of gamma and C. Still no luck..

I am wondering if this:

1. has to do with the data (training set size) or
2. am I just using the wrong estimator or
3. my code is messed-up?

Here is the complete python code:

    import pandas as pd

    training_data = pd.read_csv("Train_wo_Header.csv") #I read my training data set
    data = training_data.ix[:,[0,1,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31]] #picking up all rows expect index 2, which is my output
    target = training_data.ix[:,[2]] 


    from sklearn import svm #Code from the URL above
    clf = svm.SVC(gamma=0.001, C=100.)
    clf.fit(data,target)  

    test_data = pd.read_csv("test_wo_Header.csv") #this is my test data

    clf.predict(test_data[-10:]) #predicting the last 10 values

Here is the output:

array([7734, 7734, 7734, 7734, 7734, 7734, 7734, 7734, 7734, 7734], dtype=int64)

I even tried using LinearSVC. Still no luck. The only difference is that the predicted output is a different value (9240), but same throughout ...

Answer 1

Always giving the same output can have 2 causes :

• your model is overfitting (unbalanced dataset ?)
• you're not giving the correct data to your model

You didn't seem to convert your Pandas DataFrame to a numpy array, try

clf = svm.SVC()    
X = data.values
Y = target.values
assert len(X) == len(Y)

clf.fit(X,Y)
print clf.score(X,Y)

Do the same for your test data and try to print at least the shape of your data and one element of your arrays.