在PyStruct中拟合SSVM模型时出现IndexError

Question

I'm using the pystruct Python module for a structured learning problem in classifying posts in discussion threads, and I've run into an issue when tying to train the OneSlackSSVM for use with the LinearChainCRF . I'm following the OCR example from the docs , but can't seem to call the .fit() method on the SSVM. Here is the error I'm getting:

Traceback (most recent call last):

File "<ipython-input-47-da804d135818>", line 1, in <module>
ssvm.fit(X_train, y_train)

File "/Users/kylefth/anaconda/lib/python2.7/site-  
packages/pystruct/learners/one_slack_ssvm.py", line 429, in fit
joint_feature_gt = self.model.batch_joint_feature(X, Y)

File "/Users/kylefth/anaconda/lib/python2.7/site-       
packages/pystruct/models/base.py", line 40, in batch_joint_feature      
joint_feature_ += self.joint_feature(x, y)

File "/Users/kylefth/anaconda/lib/python2.7/site-    
packages/pystruct/models/graph_crf.py", line 197, in joint_feature
unary_marginals[gx, y] = 1

IndexError: index 7 is out of bounds for axis 1 with size 7

Below is the code I've written. I've tired to structure the data as in the docs example where the overall data structure is a dict with keys for data , labels , and folds .

from pystruct.models import LinearChainCRF
from pystruct.learners import OneSlackSSVM

# Printing out keys of overall data structure
print threads.keys()
>>> ['folds', 'labels', 'data']

# Creating instances of models
crf = LinearChainCRF()
ssvm = OneSlackSSVM(model=crf)

# Splitting up data into training and test sets as in example
X, y, folds = threads['data'], threads['labels'], threads['folds']
X_train, X_test = X[folds == 1], X[folds != 1]
y_train, y_test = y[folds == 1], y[folds != 1]

# Print out dimensions of first element in data and labels
print X[0].shape, y[0].shape
>>> (8, 211), (8,)

# Fitting the ssvm model
ssvm.fit(X_train, y_train)
>>> see error above

Directly after trying to fit the model, I get the above error. All instances of X_train , X_test , y_train , and y_test have 211 columns and all the label dimensions appear to match up with their corresponding training and testing data. Any help would be greatly appreciated.

Answer 1

I think everything you are doing is right, this is https://github.com/pystruct/pystruct/issues/114 . Your labels y need to start from 0 to n_labels. I think yours start at 1.