Creating a sentence-transformer model in Spark Mllib - apache-spark

I used a pre-trained model from the sentence transformer library to check the similarity between two sentences. Now I need this particular model to be implemented using spark mllib. Any Suggestions? I really appreciate any help you can provide.
https://www.sbert.net/
https://spark.apache.org/mllib/

One approach that I found to work is to use a Pandas UDF that encodes the text and returns the embedding. This embedding column could then be used with MLlib.
import pandas as pd
import pyspark.sql.functions as F
from pyspark.sql.types import ArrayType, DoubleType, StringType
from sentence_transformers import SentenceTransformer
# import sbert model
model = SentenceTransformer("all-MiniLM-L6-v2")
# sentences to encode
sentences = [
"This framework generates embeddings for each input sentence",
"Sentences are passed as a list of string.",
"The quick brown fox jumps over the lazy dog.",
]
# create spark df with sentences
data = spark.createDataFrame(sentences, StringType(), ["sentences"])
data.show()
# create a pandas udf that will encode the text and return an array of doubles
#F.pandas_udf(returnType=ArrayType(DoubleType()))
def encode(x: pd.Series) -> pd.Series:
return pd.Series(model.encode(x).tolist())
# apply udf and show
data.withColumn("embedding", encode("value")).show()
output
+--------------------+
| value|
+--------------------+
|This framework ge...|
|Sentences are pas...|
|The quick brown f...|
+--------------------+
+--------------------+--------------------+
| value| embedding|
+--------------------+--------------------+
|This framework ge...|[-0.0137173617258...|
|Sentences are pas...|[0.05645250156521...|
|The quick brown f...|[0.04393352568149...|
+--------------------+--------------------+

Related

Preparing data for LDA training with PySpark 1.6

I have a corpus of documents that I'm reading into a spark data frame.
I have tokeniked and vectorized the text and now I want to feed the vectorized data into an mllib LDA model. The LDA API docs seems to require the data to be:
rdd – RDD of documents, which are tuples of document IDs and term (word) count vectors. The term count vectors are “bags of words” with a fixed-size vocabulary (where the vocabulary size is the length of the vector). Document IDs must be unique and >= 0.
How can I get from my data frame to a suitable rdd?
from pyspark.mllib.clustering import LDA
from pyspark.ml.feature import Tokenizer
from pyspark.ml.feature import CountVectorizer
#read the data
tf = sc.wholeTextFiles("20_newsgroups/*")
#transform into a data frame
df = tf.toDF(schema=['file','text'])
#tokenize
tokenizer = Tokenizer(inputCol="text", outputCol="words")
tokenized = tokenizer.transform(df)
#vectorize
cv = CountVectorizer(inputCol="words", outputCol="vectors")
model = cv.fit(tokenized)
result = model.transform(tokenized)
#transform into a suitable rdd
myrdd = ?
#LDA
model = LDA.train(myrdd, k=2, seed=1)
PS : I'm using Apache Spark 1.6.3
Let's first organize imports, read the data, do some simple special characters removal and transform it into a DataFrame:
import re # needed to remove special character
from pyspark import Row
from pyspark.ml.feature import StopWordsRemover
from pyspark.ml.feature import Tokenizer, CountVectorizer
from pyspark.mllib.clustering import LDA
from pyspark.sql import functions as F
from pyspark.sql.types import StructType, StructField, LongType
pattern = re.compile('[\W_]+')
rdd = sc.wholeTextFiles("./data/20news-bydate/*/*/*") \
.mapValues(lambda x: pattern.sub(' ', x)).cache() # ref. https://stackoverflow.com/a/1277047/3415409
df = rdd.toDF(schema=['file', 'text'])
We will need to add an index to each Row. The following code snippet is inspired from this question about adding primary keys with Apache Spark :
row_with_index = Row(*["id"] + df.columns)
def make_row(columns):
def _make_row(row, uid):
row_dict = row.asDict()
return row_with_index(*[uid] + [row_dict.get(c) for c in columns])
return _make_row
f = make_row(df.columns)
indexed = (df.rdd
.zipWithUniqueId()
.map(lambda x: f(*x))
.toDF(StructType([StructField("id", LongType(), False)] + df.schema.fields)))
Once we have added the index, we can proceed to the features cleansing, extraction and transformation :
# tokenize
tokenizer = Tokenizer(inputCol="text", outputCol="tokens")
tokenized = tokenizer.transform(indexed)
# remove stop words
remover = StopWordsRemover(inputCol="tokens", outputCol="words")
cleaned = remover.transform(tokenized)
# vectorize
cv = CountVectorizer(inputCol="words", outputCol="vectors")
count_vectorizer_model = cv.fit(cleaned)
result = count_vectorizer_model.transform(cleaned)
Now, let's transform the results dataframe back to rdd
corpus = result.select(F.col('id').cast("long"), 'vectors').rdd \
.map(lambda x: [x[0], x[1]])
Our data is now ready to be trained :
# training data
lda_model = LDA.train(rdd=corpus, k=10, seed=12, maxIterations=50)
# extracting topics
topics = lda_model.describeTopics(maxTermsPerTopic=10)
# extraction vocabulary
vocabulary = count_vectorizer_model.vocabulary
We can print the topics descriptions now as followed :
for topic in range(len(topics)):
print("topic {} : ".format(topic))
words = topics[topic][0]
scores = topics[topic][1]
[print(vocabulary[words[word]], "->", scores[word]) for word in range(len(words))]
PS : This above code was tested with Spark 1.6.3.

convert dataframe to libsvm format

I have a dataframe resulting from a sql query
df1 = sqlContext.sql("select * from table_test")
I need to convert this dataframe to libsvm format so that it can be provided as an input for
pyspark.ml.classification.LogisticRegression
I tried to do the following. However, this resulted in the following error as I'm using spark 1.5.2
df1.write.format("libsvm").save("data/foo")
Failed to load class for data source: libsvm
I wanted to use MLUtils.loadLibSVMFile instead. I'm behind a firewall and can't directly pip install it. So I downloaded the file, scp-ed it and then manually installed it. Everything seemed to work fine but I still get the following error
import org.apache.spark.mllib.util.MLUtils
No module named org.apache.spark.mllib.util.MLUtils
Question 1: Is my above approach to convert dataframe to libsvm format in the right direction.
Question 2: If "yes" to question 1, how to get MLUtils working. If "no", what is the best way to convert dataframe to libsvm format
I would act like that (it's just an example with an arbitrary dataframe, I don't know how your df1 is done, focus is on data transformations):
This is my way to convert dataframe to libsvm format:
# ... your previous imports
from pyspark.mllib.util import MLUtils
from pyspark.mllib.regression import LabeledPoint
# A DATAFRAME
>>> df.show()
+---+---+---+
| _1| _2| _3|
+---+---+---+
| 1| 3| 6|
| 4| 5| 20|
| 7| 8| 8|
+---+---+---+
# FROM DATAFRAME TO RDD
>>> c = df.rdd # this command will convert your dataframe in a RDD
>>> print (c.take(3))
[Row(_1=1, _2=3, _3=6), Row(_1=4, _2=5, _3=20), Row(_1=7, _2=8, _3=8)]
# FROM RDD OF TUPLE TO A RDD OF LABELEDPOINT
>>> d = c.map(lambda line: LabeledPoint(line[0],[line[1:]])) # arbitrary mapping, it's just an example
>>> print (d.take(3))
[LabeledPoint(1.0, [3.0,6.0]), LabeledPoint(4.0, [5.0,20.0]), LabeledPoint(7.0, [8.0,8.0])]
# SAVE AS LIBSVM
>>> MLUtils.saveAsLibSVMFile(d, "/your/Path/nameFolder/")
What you will see on the "/your/Path/nameFolder/part-0000*" files is:
1.0 1:3.0 2:6.0
4.0 1:5.0 2:20.0
7.0 1:8.0 2:8.0
See here for LabeledPoint docs
I had to do this for it to work
D.map(lambda line: LabeledPoint(line[0],[line[1],line[2]]))
If you want to convert sparse vectors to a 'sparse' libsvm which is more efficient, try this:
from pyspark.ml.linalg import Vectors
from pyspark.mllib.linalg import Vectors as MLLibVectors
from pyspark.mllib.regression import LabeledPoint
from pyspark.mllib.util import MLUtils
df = spark.createDataFrame([
(0, Vectors.sparse(5, [(1, 1.0), (3, 7.0)])),
(1, Vectors.sparse(5, [(1, 1.0), (3, 7.0)])),
(1, Vectors.sparse(5, [(1, 1.0), (3, 7.0)]))
], ["label", "features"])
df.show()
# +-----+-------------------+
# |label| features|
# +-----+-------------------+
# | 0|(5,[1,3],[1.0,7.0])|
# | 1|(5,[1,3],[1.0,7.0])|
# | 1|(5,[1,3],[1.0,7.0])|
# +-----+-------------------+
MLUtils.saveAsLibSVMFile(df.rdd.map(lambda x: LabeledPoint(x.label, MLLibVectors.fromML(x.features))), './libsvm')

PySpark, Decision Trees (Spark 2.0.0)

I am new to Spark (using PySpark). I tried running the Decision Tree tutorial from here (link). I execute the code:
from pyspark.ml import Pipeline
from pyspark.ml.classification import DecisionTreeClassifier
from pyspark.ml.feature import StringIndexer, VectorIndexer
from pyspark.ml.evaluation import MulticlassClassificationEvaluator
from pyspark.mllib.util import MLUtils
# Load and parse the data file, converting it to a DataFrame.
data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt").toDF()
labelIndexer = StringIndexer(inputCol="label", outputCol="indexedLabel").fit(data)
# Now this line fails
featureIndexer =\
VectorIndexer(inputCol="features", outputCol="indexedFeatures", maxCategories=4).fit(data)
I get the error message:
IllegalArgumentException: u'requirement failed: Column features must be of type org.apache.spark.ml.linalg.VectorUDT#3bfc3ba7 but was actually org.apache.spark.mllib.linalg.VectorUDT#f71b0bce.'
When searching the web for this error I found an answer that says:
use
from pyspark.ml.linalg import Vectors, VectorUDT
instead of
from pyspark.mllib.linalg import Vectors, VectorUDT
which is odd, since I haven't used it. Also, adding this import to my code solves nothing and I still get the same error.
I am not quite clear on how to debug this situation. When looking into the raw data I see:
data.show()
+--------------------+-----+
| features|label|
+--------------------+-----+
|(692,[127,128,129...| 0.0|
|(692,[158,159,160...| 1.0|
|(692,[124,125,126...| 1.0|
|(692,[152,153,154...| 1.0|
This looks like a list, starts with '('.
I am not sure how to solve this issue, or even debug.
The source of the problem seems to be executing spark 1.5.2. example on spark 2.0.0 (see below reference to spark 2.0 example).
The difference between spark.ml and spark.mllib
As of Spark 2.0, the RDD-based APIs in the spark.mllib package have entered maintenance mode. The primary Machine Learning API for Spark is now the DataFrame-based API in the spark.ml package.
More details can be found here: http://spark.apache.org/docs/latest/ml-guide.html
Using spark 2.0 please try Spark 2.0.0 example (https://spark.apache.org/docs/2.0.0/mllib-decision-tree.html)
from pyspark.mllib.tree import DecisionTree, DecisionTreeModel
from pyspark.mllib.util import MLUtils
# Load and parse the data file into an RDD of LabeledPoint.
data = MLUtils.loadLibSVMFile(sc, 'data/mllib/sample_libsvm_data.txt')
# Split the data into training and test sets (30% held out for testing)
(trainingData, testData) = data.randomSplit([0.7, 0.3])
# Train a DecisionTree model.
# Empty categoricalFeaturesInfo indicates all features are continuous.
model = DecisionTree.trainClassifier(trainingData, numClasses=2, categoricalFeaturesInfo={},
impurity='gini', maxDepth=5, maxBins=32)
# Evaluate model on test instances and compute test error
predictions = model.predict(testData.map(lambda x: x.features))
labelsAndPredictions = testData.map(lambda lp: lp.label).zip(predictions)
testErr = labelsAndPredictions.filter(lambda (v, p): v != p).count() / float(testData.count())
print('Test Error = ' + str(testErr))
print('Learned classification tree model:')
print(model.toDebugString())
# Save and load model
model.save(sc, "target/tmp/myDecisionTreeClassificationModel")
sameModel = DecisionTreeModel.load(sc, "target/tmp/myDecisionTreeClassificationModel")
Find full example code at "examples/src/main/python/mllib/decision_tree_classification_example.py" in the Spark repo.

How to access element of a VectorUDT column in a Spark DataFrame?

I have a dataframe df with a VectorUDT column named features. How do I get an element of the column, say first element?
I've tried doing the following
from pyspark.sql.functions import udf
first_elem_udf = udf(lambda row: row.values[0])
df.select(first_elem_udf(df.features)).show()
but I get a net.razorvine.pickle.PickleException: expected zero arguments for construction of ClassDict(for numpy.dtype) error. Same error if I do first_elem_udf = first_elem_udf(lambda row: row.toArray()[0]) instead.
I also tried explode() but I get an error because it requires an array or map type.
This should be a common operation, I think.
Convert output to float:
from pyspark.sql.types import DoubleType
from pyspark.sql.functions import lit, udf
def ith_(v, i):
try:
return float(v[i])
except ValueError:
return None
ith = udf(ith_, DoubleType())
Example usage:
from pyspark.ml.linalg import Vectors
df = sc.parallelize([
(1, Vectors.dense([1, 2, 3])),
(2, Vectors.sparse(3, [1], [9]))
]).toDF(["id", "features"])
df.select(ith("features", lit(1))).show()
## +-----------------+
## |ith_(features, 1)|
## +-----------------+
## | 2.0|
## | 9.0|
## +-----------------+
Explanation:
Output values have to be reserialized to equivalent Java objects. If you want to access values (beware of SparseVectors) you should use item method:
v.values.item(0)
which return standard Python scalars. Similarly if you want to access all values as a dense structure:
v.toArray().tolist()
If you prefer using spark.sql, you can use the follow custom function 'to_array' to convert the vector to array. Then you can manipulate it as an array.
from pyspark.sql.types import ArrayType, DoubleType
def to_array_(v):
return v.toArray().tolist()
from pyspark.sql import SQLContext
sqlContext=SQLContext(spark.sparkContext, sparkSession=spark, jsqlContext=None)
sqlContext.udf.register("to_array",to_array_, ArrayType(DoubleType()))
example
from pyspark.ml.linalg import Vectors
df = sc.parallelize([
(1, Vectors.dense([1, 2, 3])),
(2, Vectors.sparse(3, [1], [9]))
]).toDF(["id", "features"])
df.createOrReplaceTempView("tb")
spark.sql("""select * , to_array(features)[1] Second from tb """).toPandas()
output
id features Second
0 1 [1.0, 2.0, 3.0] 2.0
1 2 (0.0, 9.0, 0.0) 9.0
I ran into the same problem with not being able to use explode(). One thing you can do is use VectorSlice from the pyspark.ml.feature library. Like so:
from pyspark.ml.feature import VectorSlicer
from pyspark.ml.linalg import Vectors
from pyspark.sql.types import Row
slicer = VectorSlicer(inputCol="features", outputCol="features_one", indices=[0])
output = slicer.transform(df)
output.select("features", "features_one").show()
For anyone trying to split the probability columns generated after training a PySpark ML model into usable columns. This does not use UDF or numpy. And this will only work for binary classification. Here lr_pred is the dataframe which has the predictions from the Logistic Regression Model.
prob_df1=lr_pred.withColumn("probability",lr_pred["probability"].cast("String"))
prob_df =prob_df1.withColumn('probabilityre',split(regexp_replace("probability", "^\[|\]", ""), ",")[1].cast(DoubleType()))
Since Spark 3.0.0 this can be done without using UDF.
from pyspark.ml.functions import vector_to_array
https://discuss.dizzycoding.com/how-to-split-vector-into-columns-using-pyspark/
Why is Vector[Double] is used in the results? That's not a very nice data type.

How to convert type Row into Vector to feed to the KMeans

when i try to feed df2 to kmeans i get the following error
clusters = KMeans.train(df2, 10, maxIterations=30,
runs=10, initializationMode="random")
The error i get:
Cannot convert type <class 'pyspark.sql.types.Row'> into Vector
df2 is a dataframe created as follow:
df = sqlContext.read.json("data/ALS3.json")
df2 = df.select('latitude','longitude')
df2.show()
latitude| longitude|
60.1643075| 24.9460844|
60.4686748| 22.2774728|
how can i convert this two columns to Vector and feed it to KMeans?
ML
The problem is that you missed the documentation's example, and it's pretty clear that the method train requires a DataFrame with a Vector as features.
To modify your current data's structure you can use a VectorAssembler. In your case it could be something like:
from pyspark.sql.functions import *
vectorAssembler = VectorAssembler(inputCols=["latitude", "longitude"],
outputCol="features")
# For your special case that has string instead of doubles you should cast them first.
expr = [col(c).cast("Double").alias(c)
for c in vectorAssembler.getInputCols()]
df2 = df2.select(*expr)
df = vectorAssembler.transform(df2)
Besides, you should also normalize your features using the class MinMaxScaler to obtain better results.
MLLib
In order to achieve this using MLLib you need to use a map function first, to convert all your string values into Double, and merge them together in a DenseVector.
rdd = df2.map(lambda data: Vectors.dense([float(c) for c in data]))
After this point you can train your MLlib's KMeans model using the rdd variable.
I got PySpark 2.3.1 to perform KMeans on a DataFrame as follows:
Write a list of the columns you want to include in the clustering analysis:
feat_cols = ['latitude','longitude']`
You need all of the columns to be numeric values:
expr = [col(c).cast("Double").alias(c) for c in feat_cols]
df2 = df2.select(*expr)
Create your features vector with mllib.linalg.Vectors:
from pyspark.ml.feature import VectorAssembler
assembler = VectorAssembler(inputCols=feat_cols, outputCol="features")
df3 = assembler.transform(df2).select('features')
You should normalize your features as normalization is not always required, but it rarely hurts (more about this here):
from pyspark.ml.feature import StandardScaler
scaler = StandardScaler(
inputCol="features",
outputCol="scaledFeatures",
withStd=True,
withMean=False)
scalerModel = scaler.fit(df3)
df4 = scalerModel.transform(df3).drop('features')\
.withColumnRenamed('scaledFeatures', 'features')
Turn your DataFrame object df4 into a dense vector RDD:
from pyspark.mllib.linalg import Vectors
data5 = df4.rdd.map(lambda row: Vectors.dense([x for x in row['features']]))
Use the obtained RDD object as input for KMeans training:
from pyspark.mllib.clustering import KMeans
model = KMeans.train(data5, k=3, maxIterations=10)
Example: classify a point p in your vector space:
prediction = model.predict(p)

Resources