This repo is a fork from the work contained in this project.
This repo aims to only support python +3.7 and pyspark +2.
- Histogram incorrectly counts Null values as zero
- Static Typing
- Add Project to pip
- Readme docs & document code
- Testing
30-11-2020:
BUG FIX: Base profiler was not counting zeros and missing values correctly with spark 2.4 (unsure about earlier versions) because `count(col(colname)) == 0.0`) does not count the filtered values but in fact counts all values.
Generates profile reports from an Apache Spark DataFrame. It is based on pandas_profiling, but for Spark's DataFrames instead of pandas'.
For each column the following statistics - if relevant for the column type - are presented in an interactive HTML report:
- Essentials: type, unique values, missing values
- Quantile statistics like minimum value, Q1, median, Q3, maximum, range, interquartile range
- Descriptive statistics like mean, mode, standard deviation, sum, median absolute deviation, coefficient of variation, kurtosis, skewness
- Most frequent values
- Histogram
All operations are done efficiently, which means that no Python UDFs or .map() transformations are used at all; only Spark SQL's Catalyst (and the Tungsten execution engine) is used for the retrieval of all statistics.
Available here.
If you are using Anaconda, you already have all the needed dependencies. So you just have to pip install the package without dependencies (just in case pip tries to overwrite your current dependencies):
pip install --no-deps spark-df-profiling
If you don't have pandas and/or matplotlib installed:
pip install spark-df-profiling
The profile report is written in HTML5 and CSS3, which means that you may require a modern browser.
Keep in mind that you need a working Spark cluster (or a local Spark installation). The report must be created from pyspark. To point pyspark driver to your Python environment, you must set the environment variable PYSPARK_DRIVER_PYTHON to your python environment where spark-df-profiling is installed. For example, for Anaconda:
export PYSPARK_DRIVER_PYTHON=/path/to/your/anaconda/bin/python
And then you can execute /path/to/your/bin/pyspark to enter pyspark's CLI.
We recommend generating reports interactively by using the Jupyter notebook.
To use pyspark with Jupyter, you must also set PYSPARK_DRIVER_PYTHON:
export PYSPARK_DRIVER_PYTHON=/path/to/your/anaconda/bin/python
And then:
IPYTHON_OPTS="notebook" /path/to/your/bin/pyspark
In spark 2.0.X IPYTHON_OPTS is removed: the environment variable you want to set is PYSPARK_DRIVER_PYTHON_OPTS:
PYSPARK_DRIVER_PYTHON_OPTS="notebook" /path/to/your/bin/pyspark
Now you can create a new notebook, which will run pyspark.
To use spark-df-profiling, start by loading in your Spark DataFrame, e.g. by using
# sqlContext is probably already created for you.
# To load a parquet file as a Spark Dataframe, you can:
df = sqlContext.read.parquet("/path/to/your/file.parquet")
# And you probably want to cache it, since a lot of
# operations will be done while the report is being generated:
df_spark = df.cache()To display the report in a Jupyter notebook, run:
import spark_df_profiling
spark_df_profiling.ProfileReport(df_spark)If you want to generate a HTML report file, save the ProfileReport to an object and use the .to_file() method:
profile = spark_df_profiling.ProfileReport(df_spark)
profile.to_file(outputfile="/tmp/myoutputfile.html")- Python (
>=3.7) - Apache Spark -> requires Spark
>=2 - An internet connection. spark-df-profiling requires an internet connection to download the Bootstrap and JQuery libraries. You can choose to embed them in the HTML template code, should you desire.
- jinja2 (
>=2.8) -> needed for template rendering. Only needed in the Spark driver. - matplotlib (
>=1.4) -> needed for histogram creation. Only needed in the Spark driver. - pandas (
>=0.17.0) -> needed for internal data arrangement. Only needed in the Spark driver. - six (
>=1.9.0) -> needed for py2/3 compatibility. Only needed in the Spark driver.