PyNotes

Transformers

My notebooks on using Tranformer models off-line for fine-tuning and prediction purposes.

May the Spark ⭐ be with you

My notebook on using Python with Jupyter Notebook, PySpark and other well known machine learning frameworks.

SparkNLP

Write a PySpark Array of Strings as String into ONE Parquet File

Use Case

You can use a PySpark Tokenizer to convert a string into tokens and apply machine learning algorithms on it. The code snippets below might be useful if you want to inspect the result of the tokenizer (an array of unicode strings) via csv file (saved in a Parquet environment).

Code

df.select("words").show()

+--------------------+
|               words|
+--------------------+
| [I, am, looking,...|
|     [not, today,...|
|   [but, tomorrow...|
+--------------------+

# Select column with array of words into seperate DataFrame
dfSave = df.select("words")

#dfSave.printSchema()
# root
#  |-- words: array (nullable = true)
#  |    |-- element: string (containsNull = true)
#

import pyspark.sql.functions as F

# Convert Array of unicode strings into a string using PySpark's function
# https://stackoverflow.com/questions/38924762/how-to-convert-column-of-arrays-of-strings-to-strings
dfSave = dfSave.withColumn("words_str", F.concat_ws(" ", dfSave["words"]))

# drop arrays of strings column
dfSave = dfSave.drop("words")

# Write dataframe with string into ONE parquet file
# https://stackoverflow.com/questions/42022890/how-can-i-write-a-parquet-file-using-spark-pyspark
# https://stackoverflow.com/questions/36162055/pyspark-spit-out-single-file-when-writing-instead-of-multiple-part-files
dfSave.coalesce(1).write.format('csv').save('/home/me/tokenized.csv')