- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content

Hello,

I am trying to use the pyDAAL SVM to do binary classification, but I keep running into an system error at the prediction compute().

I wrote a small reproducer for the issue I was seeing on my original code:

import numpy as np from daal.data_management import HomogenNumericTable from daal.algorithms.svm import training, prediction from daal.algorithms import kernel_function, classifier import daal.algorithms.kernel_function.rbf import daal.algorithms.classifier.training import daal.algorithms.classifier.prediction trn_X = HomogenNumericTable(np.random.randn(3000,10)) trn_Y = HomogenNumericTable(np.random.randn(3000,1)) tst_X = HomogenNumericTable(np.random.randn(1000,10)) kernel = kernel_function.rbf.Batch() trainer = training.Batch() trainer.parameter.kernel = kernel trainer.parameter.C = 1.0 trainer.parameter.gamma = 0.001 trainer.input.set(classifier.training.data, trn_X) trainer.input.set(classifier.training.labels, trn_Y) trn_res = trainer.compute() predictor = prediction.Batch() predictor.parameter.kernel = kernel predictor.input.setTable(classifier.prediction.data, tst_X) predictor.input.setModel(classifier.prediction.model, trn_res.get(classifier.training.model)) predictor.compute()

And running this reproducer gives me this error:

Traceback (most recent call last): File "svm-reproducer.py", line 27, in <module> predictor.compute() File "/opt/intel/intelpython_update_3/intelpython2/lib/python2.7/site-packages/daal/algorithms/svm/prediction.py", line 185, in compute return _prediction12.Batch_Float64DefaultDense_compute(self) SystemError: Model is not full initialized Details: Argument name: supportVectors

Is there something else that I am missing? I took a look at the example directory, and the online github examples, but I could not figure out what the step I am missing is. As shown in the error, I am using 2017 update 3 pyDAAL.

Thanks in advance,

Ryo Asai

Link Copied

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content

Hello Ryo,

In your example SVM model is not trained due to incorrect values of the class labels.

trn_Y numeric table should contain values {-1, 1} as specified in the DAAL Developer Guide, "Support Vector Machine Classifier > Details":

“Given `n` feature vectors `x` _{1}= (`x` _{11},…,`x` _{1p }), ..., `x` ` _{n}` = (

`x`

_{n}_{1},…,

`x`

`) of size`

_{np}`p`and a vector of

__class labels y = (__describes the class to which the feature vector

`y`_{1},…,`y``), where`_{n}`y``∈ {-1,1}`_{i}`x`

`belongs”.`

_{i}The library does not check the content input data for validity as scanning the whole numeric tables will affect the performance of the computations.The library’s documentation such as Intel DAAL Developer Guide specifies the requirements to the parameters of the algorithms and their representation.

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content

Ah I see. I was able to train the model for both the code snippet and the original code.

Thank you so much for your help!

Ryo

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page