Sloane Simmons created SPARK-12680:
--------------------------------------
Summary: Loading Word2Vec model in pyspark gives "ValueError: too
many values to unpack" in findSynonyms
Key: SPARK-12680
URL: https://issues.apache.org/jira/browse/SPARK-12680
Project: Spark
Issue Type: Bug
Components: MLlib, PySpark
Affects Versions: 1.5.2
Reporter: Sloane Simmons
I can train a model with Word2Vec and then persist it with Word2VecModel#save.
If I load the saved model in pyspark (using python 2.7.10), I get the following
error (model.transform included to show that other methods work).
{code}
In [3]: from pyspark.mllib.feature import Word2VecModel
In [4]: model = Word2VecModel.load(sc,"word_vec_from_cleaned_query.model")
SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further
details.
16/01/06 12:36:11 WARN BLAS: Failed to load implementation from:
com.github.fommil.netlib.NativeSystemBLAS
16/01/06 12:36:11 WARN BLAS: Failed to load implementation from:
com.github.fommil.netlib.NativeRefBLAS
In [5]: model.findSynonyms('white',10)
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-5-6105c004edd9> in <module>()
----> 1 model.findSynonyms('white',10)
/usr/local/Cellar/apache-spark/1.5.2/libexec/python/pyspark/mllib/feature.pyc
in findSynonyms(self, word, num)
448 if not isinstance(word, basestring):
449 word = _convert_to_vector(word)
--> 450 words, similarity = self.call("findSynonyms", word, num)
451 return zip(words, similarity)
452
ValueError: too many values to unpack
In [6]: model.transform('white')
Out[6]: DenseVector([-0.0213, 0.2292, -0.2012, 0.107, -0.1475, 0.0578, 0.0731,
-0.098, -0.1528, 0.1077, 0.0158, -0.0155, -0.1487, 0.0343, 0.2244, 0.0447,
0.2362, -0.1767, 0.064, -0.0148, -0.1291, -0.0171, -0.0642, -0.0754, 0.0417,
0.1547, 0.2745, -0.1178, -0.2895, -0.1314, 0.1023, -0.11, 0.0142, 0.0156,
0.1102, 0.0785, -0.0981, 0.0504, -0.0627, -0.0773, 0.0023, 0.1826, 0.1759,
-0.1581, 0.3913, 0.0829, 0.0728, 0.1478, -0.0123, -0.1745, 0.2762, 0.0312,
0.138, 0.0786, -0.0546, 0.5123, 0.237, -0.0241, 0.1594, -0.0645, -0.0425,
0.1265, 0.0305, -0.3164, 0.0601, 0.0565, 0.0066, -0.0818, -0.384, -0.1513,
0.0775, -0.2278, -0.1478, -0.0659, -0.0778, 0.3194, -0.1931, -0.2991, 0.1629,
0.1018, -0.0603, 0.1091, -0.0334, -0.0513, 0.1067, 0.1273, 0.1187, 0.0461,
0.0407, 0.0515, 0.0958, 0.0498, -0.1561, 0.1726, -0.006, -0.0262, -0.0106,
0.1623, 0.1477, -0.0509])
In [7]:
{code}
I think that this is a pyspark-specific error, since I can load the trained
model in the scala spark-shell and use findSynonyms:
{code}
scala> import org.apache.spark.mllib.feature.{Word2Vec, Word2VecModel}
import org.apache.spark.mllib.feature.{Word2Vec, Word2VecModel}
scala> val model = Word2VecModel.load(sc,"word_vec_from_cleaned_query.model")
SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further
details.
16/01/06 14:17:14 WARN BLAS: Failed to load implementation from:
com.github.fommil.netlib.NativeSystemBLAS
16/01/06 14:17:14 WARN BLAS: Failed to load implementation from:
com.github.fommil.netlib.NativeRefBLAS
model: org.apache.spark.mllib.feature.Word2VecModel =
org.apache.spark.mllib.feature.Word2VecModel@2e7da886
scala> model.findSynonyms("white",10)
res0: Array[(String, Double)] = Array((stylish,0.8347662648041679),
(shirt,0.7721922530954246), (stripe,0.7311193884955149),
(striped,0.7033047124091971), (buttons,0.6891310548525095),
(womens,0.671437501511924), (zaful,0.6659281321485323),
(dorateymur,0.6654344754707424), (womenns,0.6637001786899768),
(long,0.6573707323598634))
scala>
{code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
