Query: What is the difference between Dask and Modin? #515
Hi @jithurjacob, thanks for the question! I have heard this a lot, so it is not silly at all!
I will give a high level summary of the main points, we are planning to discuss this more in future blog posts. The TL;DR is that Modin's API is identical to pandas, whereas Dask's is not.
Note: The projects are fundamentally different in their aims, so a fair comparison is challenging.
Also, I will assume that you are asking about Dask DataFrame, rather than Dask as a whole. Dask's compute engine is more appropriately compared to Ray, which this project uses. See a comparison of Ray vs Dask here.
Dask DataFrame does not scale the entire pandas API, and it isn't trying to. See this explained in their documentation here: http://docs.dask.org/en/latest/dataframe.html#common-uses-and-anti-uses
Dask DataFrames API is also different from the pandas API in that it is lazy and needs .compute() to materialize the DataFrame. This makes the API less convenient but allows to do certain query optimizations/rearrangement, which can give speedups in certain situations. We are planning to incorporate similar capabilities into Modin but hope we can do so without having to change the API. We will outline plans for speeding up Modin in an upcoming blog post.
Modin attempts to parallelize as much of the pandas API as is possible. We have worked through a significant portion of the DataFrame API. It is intended to be used as a drop-in replacement for pandas, such that even if the API is not yet parallelized, it is still defaulting to pandas.
Dask DataFrame has row-based partitioning, similar to Spark. This can be seen in their documentation: http://docs.dask.org/en/latest/dataframe.html#design. They also have a custom index object for indexing into the object, which is not pandas compatible. Dask DataFrame seems to treat operations on the DataFrame as MapReduce operations, which is a good paradigm for the subset of the pandas API they have chosen to implement.
Modin is more of a column-store, which we inherited from modern database systems. We laterally partition the columns for scalability (many systems, such as Google BigTable already did this), so we can scale in both directions and have finer grained partitioning. This is explained at a high level in Modin's documentation: https://modin.readthedocs.io/en/latest/architecture.html. Because we have this finer grained control over the partitioning, we can support a number of operations that are very challenging in MapReduce systems (e.g. transpose, median, quantile).
In the long-term, Modin is planned to become a DataFrame library that supports the popular APIs (SQL, pandas, etc.) and runs on a variety of compute engines and backends. In fact, a group was able to contribute a
edit: Add tl;dr above