Spark is a unified analytics engine for large-scale data processing. It provides high-level APIs in Scala, Java, Python, and R, and an optimized engine that supports general computation graphs for data analysis. It also supports a rich set of higher-level tools including Spark SQL for SQL and DataFrames, MLlib for machine learning, GraphX for graph processing, and Structured Streaming for stream processing.
Spark MLlib library for Machine Learning provides a Collaborative Filtering implementation by using Alternating Least Squares. The implementation in MLlib has these parameters:
- numBlocks is the number of blocks used to parallelize computation (set to -1 to auto-configure).
- rank is the number of latent factors in the model.
- iterations is the number of iterations to run.
- lambda specifies the regularization parameter in ALS.
- implicitPrefs specifies whether to use the explicit feedback ALS variant or one adapted for implicit feedback data.
- alpha is a parameter applicable to the implicit feedback variant of ALS that governs the baseline confidence in preference observations.
Often data sources are incomplete, which means you will have missing data, you have 3 basic options for filling in missing data (you will personally have to make the decision for what is the right approach:
- Just keep the missing data points.
- Drop them missing data points (including the entire row)
- Fill them in with some other value.
- Linear regression
- Logistic regression
- Natural Language Processing (NLP)
- Tree methods
- Recomender System
You won't always get data in a convienent format, often you will have to deal with data that is non-numerical, such as customer names, or zipcodes, country names, etc...
A big part of working with data is using your own domain knowledge to build an intuition of how to deal with the data, sometimes the best course of action is to drop the data, other times feature-engineering is a good way to go, or you could try to transform the data into something the Machine Learning Algorithms will understand.
Spark has several built in methods of dealing with thse transformations, check them all out here: http://spark.apache.org/docs/latest/ml-features.html
VectorAssembler is a transformer that combines a given list of columns into a single vector column. It is useful for combining raw features and features generated by different feature transformers into a single feature vector, in order to train ML models like logistic regression and decision trees. VectorAssembler accepts the following input column types: all numeric types, boolean type, and vector type. In each row, the values of the input columns will be concatenated into a vector in the specified order.
Assume that we have a DataFrame with the columns id, hour, mobile, userFeatures, and clicked:
id | hour | mobile | userFeatures | clicked
----|------|--------|------------------|---------
0 | 18 | 1.0 | [0.0, 10.0, 0.5] | 1.0
userFeatures is a vector column that contains three user features. We want to combine hour, mobile, and userFeatures into a single feature vector called features and use it to predict clicked or not. If we set VectorAssembler’s input columns to hour, mobile, and userFeatures and output column to features, after transformation we should get the following DataFrame:
id | hour | mobile | userFeatures | clicked | features
----|------|--------|------------------|---------|-----------------------------
0 | 18 | 1.0 | [0.0, 10.0, 0.5] | 1.0 | [18.0, 1.0, 0.0, 10.0, 0.5]
The image below (from Wikipedia) shows an example of collaborative filtering. At first, people rate different items (like videos, images, games). Then, the system makes predictions about a user's rating for an item not rated yet. The new predictions are built upon the existing ratings of other users with similar ratings with the active user. In the image, the system predicts that the user will not like the video.
With Collaborative filtering we make predictions (filtering) about the interests of a user by collecting preferences or taste information from many users (collaborating). The underlying assumption is that if a user A has the same opinion as a user B on an issue, A is more likely to have B's opinion on a different issue x than to have the opinion on x of a user chosen randomly.
Streaming is something that is rapidly advancing and changin fast, there are multipl enew libraries every year, new and different services always popping up, and what is in this notebook may or may not apply to you. Maybe your looking for something specific on Kafka, or maybe you are looking for streaming about twitter, in which case Spark might be overkill for what you really want. Realistically speaking each situation is going to require a customized solution and this course is never going to be able to supply a one size fits all solution.
Spark has pretty well known Streaming Capabilities, if streaming is something you've found yourself needing at work then you are probably familiar with some of these concepts already, in which case you may find it more useful to jump straight to the official documentation here:
It is really a great guide, but keep in mind some of the features are restricted to Scala at this time (Spark 2.1), hopefully they will be expanded to the Python API in the future!
Twitter is a great source for streaming because its something most people already have an intuitive understanding of, you can visit the site yourself, and a lot of streaming technology has come out of Twitter as a company. You don't access to the entire "firehose" of twitter without paying for it, but that would be too much for us to handle anyway, so we'll be more than fine with the freely available API access.
Spark Streaming is an extension of the core Spark API that enables scalable, high-throughput, fault-tolerant stream processing of live data streams. Data can be ingested from many sources like Kafka, Flume, Kinesis, or TCP sockets, and can be processed using complex algorithms expressed with high-level functions like map, reduce, join and window. Finally, processed data can be pushed out to filesystems, databases, and live dashboards. In fact, you can apply Spark’s machine learning and graph processing algorithms on data streams.
Keep in mind that a few of these Streamiing Capabilities are limited when it comes to Python, you'll need to reference the documentation for the most up to date information. Also the streaming contexts tend to follow more along with the older RDD syntax, so a few things might seem different than what we are used to seeing, keep that in mind, you'll definitely want to have a good understanding of lambda expressions before continuing with this!
There are SparkSQL modules for streaming:
But they are all still listed as experimental, so instead of showing you somethign that might break in the future, we'll stick to the RDD methods (which is what the documentation also currently shows for streaming).
Internally, it works as follows. Spark Streaming receives live input data streams and divides the data into batches, which are then processed by the Spark engine to generate the final stream of results in batches.
