Extensions
In this notebook, we introduce multiple improvements to the original algorithm, using the small dataset. We: - First, improve the original algorithm by creating a system that takes user info and outputs suggestions, which is the typical final role of a recommendation system. - Then, we add the functionality that for first time user, we output the top rated movies over all users. - Furthemore, we improve the existing model by including the movie's genres to give better recommendations.
// import the relevant libraries for `mllib`
import org.apache.spark.mllib.recommendation.ALS
import org.apache.spark.mllib.recommendation.MatrixFactorizationModel
import org.apache.spark.mllib.recommendation.Rating
import org.apache.spark.sql.expressions.UserDefinedFunction
import scala.collection.mutable.WrappedArray
import org.apache.spark.mllib.recommendation.ALS
import org.apache.spark.mllib.recommendation.MatrixFactorizationModel
import org.apache.spark.mllib.recommendation.Rating
import org.apache.spark.sql.expressions.UserDefinedFunction
import scala.collection.mutable.WrappedArray
// Get the small dataset and display information
display(dbutils.fs.ls("/databricks-datasets/cs100/lab4/data-001/")) // The data is already here
| path | name | size |
|---|---|---|
| dbfs:/databricks-datasets/cs100/lab4/data-001/movies.dat | movies.dat | 171308.0 |
| dbfs:/databricks-datasets/cs100/lab4/data-001/ratings.dat.gz | ratings.dat.gz | 2837683.0 |
// Create the RDD containing the ratings
val ratingsRDD = sc.textFile("/databricks-datasets/cs100/lab4/data-001/ratings.dat.gz").map { line =>
val fields = line.split("::")
// format: Rating(userId, movieId, rating)
Rating(fields(0).toInt, fields(1).toInt, fields(2).toDouble)
}
ratingsRDD: org.apache.spark.rdd.RDD[org.apache.spark.mllib.recommendation.Rating] = MapPartitionsRDD[21729] at map at command-3389902380791579:3
// We take a look at the first 10 entries in the Ratings RDD
ratingsRDD.take(10).map(println)
Rating(1,1193,5.0)
Rating(1,661,3.0)
Rating(1,914,3.0)
Rating(1,3408,4.0)
Rating(1,2355,5.0)
Rating(1,1197,3.0)
Rating(1,1287,5.0)
Rating(1,2804,5.0)
Rating(1,594,4.0)
Rating(1,919,4.0)
res86: Array[Unit] = Array((), (), (), (), (), (), (), (), (), ())
A similar command is used to format the movies. For this first part the genre field is ignored. They will considered in the second part of this notebook.
val movies = sc.textFile("/databricks-datasets/cs100/lab4/data-001/movies.dat").map { line =>
val fields = line.split("::")
// format: (movieId, movieName)
(fields(0).toInt, fields(1))
}.collect.toMap
// check the size of the small dataset
val numRatings = ratingsRDD.count
val numUsers = ratingsRDD.map(_.user).distinct.count
val numMovies = ratingsRDD.map(_.product).distinct.count
println("Got " + numRatings + " ratings from "
+ numUsers + " users on " + numMovies + " movies.")
Got 487650 ratings from 2999 users on 3615 movies.
numRatings: Long = 487650
numUsers: Long = 2999
numMovies: Long = 3615
// Creating a Training Set, test Set and Validation Set
val Array(trainingRDD, validationRDD, testRDD) = ratingsRDD.randomSplit(Array(0.60, 0.20, 0.20), 0L)
trainingRDD: org.apache.spark.rdd.RDD[org.apache.spark.mllib.recommendation.Rating] = MapPartitionsRDD[21741] at randomSplit at command-3389902380791584:3
validationRDD: org.apache.spark.rdd.RDD[org.apache.spark.mllib.recommendation.Rating] = MapPartitionsRDD[21742] at randomSplit at command-3389902380791584:3
testRDD: org.apache.spark.rdd.RDD[org.apache.spark.mllib.recommendation.Rating] = MapPartitionsRDD[21743] at randomSplit at command-3389902380791584:3
// let's find the exact sizes we have next
println(" training data size = " + trainingRDD.count() +
", validation data size = " + validationRDD.count() +
", test data size = " + testRDD.count() + ".")
training data size = 292318, validation data size = 97175, test data size = 98157.
Ratings distribution
For curiosity, we start by plotting the histogram of the ratings present in this dataset
// Create a DataFrame with the data
import org.apache.spark.sql.functions._
val ratingsDF = sc.textFile("/databricks-datasets/cs100/lab4/data-001/ratings.dat.gz").map { line =>
val fields = line.split("::")
// format: (timestamp % 10, Rating(userId, movieId, rating))
(fields(0).toInt, fields(1).toInt, fields(2).toDouble)
}.toDF("userID", "movieID", "rating")
display(ratingsDF)
val history = ratingsDF.groupBy("rating").count().orderBy(asc("rating"))
history.show()
+------+------+
|rating| count|
+------+------+
| 1.0| 27472|
| 2.0| 53838|
| 3.0|127216|
| 4.0|170579|
| 5.0|108545|
+------+------+
history: org.apache.spark.sql.Dataset[org.apache.spark.sql.Row] = [rating: double, count: bigint]
display(history)
Create a system that takes user info and outputs suggestions.
user info = ((movieID,rating),(movieID,rating)). It is basically an (incomplete) line in the ratings matrix.
- Choose an user
- Run the model and fill the columns - predict the ratings for the movies
- Output the ones with the best predicted score
// Train the model as usually
val rank = 4
val numIterations = 10
val regularizationParameter = 0.01
val model = ALS.train(trainingRDD, rank, numIterations, regularizationParameter)
rank: Int = 4
numIterations: Int = 10
regularizationParameter: Double = 0.01
model: org.apache.spark.mllib.recommendation.MatrixFactorizationModel = org.apache.spark.mllib.recommendation.MatrixFactorizationModel@570cf4b3
// Choose any random user,which is going to be our test user
val newUserID = 1000
// Create a list with the MovieIds we want to predict its rating to
val newUser = Array(Rating(newUserID, 1, 0),Rating(newUserID, 2, 0),Rating(newUserID.toInt, 3, 0),Rating(newUserID.toInt, 4, 0),Rating(newUserID.toInt, 5, 0))
newUser.map(println)
// Convert it to an RDD
val newTest = sc.parallelize(newUser)
newTest.map(println)
Rating(1000,1,0.0)
Rating(1000,2,0.0)
Rating(1000,3,0.0)
Rating(1000,4,0.0)
Rating(1000,5,0.0)
newUserID: Int = 1000
newUser: Array[org.apache.spark.mllib.recommendation.Rating] = Array(Rating(1000,1,0.0), Rating(1000,2,0.0), Rating(1000,3,0.0), Rating(1000,4,0.0), Rating(1000,5,0.0))
newTest: org.apache.spark.rdd.RDD[org.apache.spark.mllib.recommendation.Rating] = ParallelCollectionRDD[21968] at parallelize at command-3389902380791591:9
res94: org.apache.spark.rdd.RDD[Unit] = MapPartitionsRDD[21969] at map at command-3389902380791591:10
// Evaluate the model on this test user
val usersProductsTest = newTest.map { case Rating(user, product, rate) =>
(user, product)
}
usersProductsTest: org.apache.spark.rdd.RDD[(Int, Int)] = MapPartitionsRDD[21970] at map at command-3389902380791592:2
// get the predictions for this test user
val predictions = model.predict(usersProductsTest)
.map { case Rating(user, product, rate)
=> ((user, product), rate)
}
val ratesAndPreds = newTest.map { case Rating(user, product, rate)
=> ((user, product), rate)
}.join(predictions)
predictions: org.apache.spark.rdd.RDD[((Int, Int), Double)] = MapPartitionsRDD[21979] at map at command-3389902380791593:3
ratesAndPreds: org.apache.spark.rdd.RDD[((Int, Int), (Double, Double))] = MapPartitionsRDD[21983] at join at command-3389902380791593:9
// Convert the RDD with the predictions to a DataFrame
val preds2 = ratesAndPreds.map { case ((user, product), (r1, r2)) => (user,product,r2) }
var predsDF = preds2.toDF("userID","movieID","pred")
predsDF.orderBy(asc("movieID"))show()
+------+-------+------------------+
|userID|movieID| pred|
+------+-------+------------------+
| 1000| 1|4.3134486083287396|
| 1000| 2| 3.561695001470941|
| 1000| 3| 3.251747295342854|
| 1000| 4|2.9727526635707116|
| 1000| 5|3.1890542732727987|
+------+-------+------------------+
preds2: org.apache.spark.rdd.RDD[(Int, Int, Double)] = MapPartitionsRDD[21984] at map at command-3389902380791594:2
predsDF: org.apache.spark.sql.DataFrame = [userID: int, movieID: int ... 1 more field]
// Order the movies according to the predictions
val orderedPreds = predsDF.orderBy(desc("pred"))
orderedPreds.show()
+------+-------+------------------+
|userID|movieID| pred|
+------+-------+------------------+
| 1000| 1|4.3134486083287396|
| 1000| 2| 3.561695001470941|
| 1000| 3| 3.251747295342854|
| 1000| 5|3.1890542732727987|
| 1000| 4|2.9727526635707116|
+------+-------+------------------+
orderedPreds: org.apache.spark.sql.Dataset[org.apache.spark.sql.Row] = [userID: int, movieID: int ... 1 more field]
// Return the ID of the highest recommended one
val t = orderedPreds.select("movieID").collect().map(_(0)).toList.take(1)
println("The movie highest recommended for this user is:")
println(movies(t(0).asInstanceOf[Int]))
The movie highest recommended for this user is:
Toy Story (1995)
t: List[Any] = List(1)
For first time users, the program gives the top rated movies over all users.
If newUser: - Check the ratings matrix - Compute the average rating of each column (of each movie) - Return the columns with the highest
// Note: This is only for the ones they said. Doesnt include the ones computed by our model...
import org.apache.spark.sql.functions._
val newUserID = 4000
// Compute the average of each movie
val averageRates = ratingsDF.groupBy("movieID").avg("rating")
averageRates.show()
+-------+------------------+
|movieID| avg(rating)|
+-------+------------------+
| 1580|3.7045454545454546|
| 2366| 3.71875|
| 1088| 3.29595015576324|
| 1959|3.6577181208053693|
| 3175|3.8145454545454545|
| 1645| 3.367021276595745|
| 496|3.3846153846153846|
| 2142|2.8256880733944953|
| 1591|2.5783132530120483|
| 2122|2.3434343434343434|
| 833| 2.130434782608696|
| 463|2.7222222222222223|
| 471| 3.665492957746479|
| 1342|2.8188976377952755|
| 148| 2.857142857142857|
| 3918| 2.806896551724138|
| 3794| 3.4|
| 1238|3.9526627218934913|
| 2866|3.7386363636363638|
| 3749| 4.0|
+-------+------------------+
only showing top 20 rows
import org.apache.spark.sql.functions._
newUserID: Int = 4000
averageRates: org.apache.spark.sql.DataFrame = [movieID: int, avg(rating): double]
// Order the movies by top ratings
val orderedRates = averageRates.orderBy(desc("avg(rating)")).withColumnRenamed("avg(rating)","avg_rate")
orderedRates.show()
+-------+-----------------+
|movieID| avg_rate|
+-------+-----------------+
| 854| 5.0|
| 853| 5.0|
| 787| 5.0|
| 1830| 5.0|
| 3881| 5.0|
| 557| 5.0|
| 3280| 5.0|
| 578| 5.0|
| 2444| 5.0|
| 3636| 5.0|
| 3443| 5.0|
| 3800| 5.0|
| 989| 5.0|
| 1002|4.666666666666667|
| 3232|4.666666666666667|
| 2839|4.666666666666667|
| 3245|4.666666666666667|
| 2905|4.609756097560975|
| 1743| 4.6|
| 2019|4.586330935251799|
+-------+-----------------+
only showing top 20 rows
orderedRates: org.apache.spark.sql.DataFrame = [movieID: int, avg_rate: double]
// Return the top 5 movies with highest ratings over all users
val topMovies = orderedRates.take(5)
//println(topMovies)
//topMovies.foreach(t => println(t(0)))
val moviesList = orderedRates.select("movieID").collect().map(_(0)).toList.take(5)
//println(moviesList)
println("The movies recommended for a new user based on the overall rating are:")
for (t <- moviesList )
println(movies(t.asInstanceOf[Int]))
// println(movies(t))
The movies recommended for a new user based on the overall rating are:
Dingo (1992)
Gate of Heavenly Peace, The (1995)
Hour of the Pig, The (1993)
Those Who Love Me Can Take the Train (Ceux qui m'aiment prendront le train) (1998)
Schlafes Bruder (Brother of Sleep) (1995)
topMovies: Array[org.apache.spark.sql.Row] = Array([989,5.0], [787,5.0], [853,5.0], [578,5.0], [3636,5.0])
moviesList: List[Any] = List(853, 787, 578, 3636, 989)
// In alternative, return the top movies with rating of 5 over all users
val topMovies5 = orderedRates.where("avg_rate == 5").select("movieID").collect().map(_(0)).toList
println("The movies recommended for a new user based on the overall rating are:")
for (t <- topMovies5 )
println(movies(t.asInstanceOf[Int]))
// println(movies(t))
The movies recommended for a new user based on the overall rating are:
Dingo (1992)
Gate of Heavenly Peace, The (1995)
Hour of the Pig, The (1993)
Those Who Love Me Can Take the Train (Ceux qui m'aiment prendront le train) (1998)
Schlafes Bruder (Brother of Sleep) (1995)
Ballad of Narayama, The (Narayama Bushiko) (1958)
Baby, The (1973)
24 7: Twenty Four Seven (1997)
Born American (1986)
Criminal Lovers (Les Amants Criminels) (1999)
Follow the Bitch (1998)
Bittersweet Motel (2000)
Mamma Roma (1962)
topMovies5: List[Any] = List(853, 787, 578, 3636, 989, 854, 3280, 2444, 3443, 3800, 1830, 3881, 557)
Genres analysis
we investigate whether suggestion based on genre can be more accurate. Imagine a scenario in which an user is interested in watching a movie of a particular genre, say an Animation movie, given this information, can we suggest a better film with respect to the film that we would have suggested by only knowing user’s previous ratings on such movie?
// Read the movies file as a dataframe and display it
val movies_df = sc.textFile("/databricks-datasets/cs100/lab4/data-001/movies.dat").map { line =>
val fields = line.split("::")
// format: (movieId, movieName,genre)
(fields(0).toInt, fields(1),fields(2).split("\\|"))
}.toDF("movieId", "movieName", "genre")
display(movies_df)
// Select a GENRE, or a set of GENREs and filter the movies dataset according to this genre
val GENRE = "Animation"
def array_contains_any(s:Seq[String]): UserDefinedFunction = {
udf((c: WrappedArray[String]) =>
c.toList.intersect(s).nonEmpty)}
val b: Array[String] = Array(GENRE)
val genre_df = movies_df.where(array_contains_any(b)($"genre"))
display(genre_df)
val movie_ID_genres = genre_df.select("movieId").rdd.map(r => r(0)).collect()
// We now read and display the ratings dataframe (without the timestamp field) as a dataframe.
val RatingsDF = sc.textFile("/databricks-datasets/cs100/lab4/data-001/ratings.dat.gz").map { line =>
val fields = line.split("::")
// format: (timestamp % 10, Rating(userId, movieId, rating))
(fields(0).toInt, fields(1).toInt, fields(2).toDouble)
}.toDF("userId", "movieId", "rating")
display(RatingsDF)
// Based on the movies id obtained by the filtering on the movie dataset we filter the ratings df and we convert it to rdd format
val Ratings_genre_df = RatingsDF.filter($"movieId".isin(movie_ID_genres:_*))
val genre_rdd = Ratings_genre_df.rdd
display(Ratings_genre_df)
// Print some dataset statistics
val numRatings = genre_rdd.count
println("Got " + numRatings + " ratings")
Got 22080 ratings
numRatings: Long = 22080
// Create train, test, and evaluation dataset and print some statistics
val Array(temp_trainingRDD, temp_validationRDD, temp_testRDD) = genre_rdd.randomSplit(Array(0.60, 0.20, 0.20), 0L)
// let's find the exact sizes we have next
println("training data size = " + temp_trainingRDD.count() +
", validation data size = " + temp_validationRDD.count() +
", test data size = " + temp_testRDD.count() + ".")
training data size = 13229, validation data size = 4411, test data size = 4440.
temp_trainingRDD: org.apache.spark.rdd.RDD[org.apache.spark.sql.Row] = MapPartitionsRDD[19929] at randomSplit at command-3389902380791621:2
temp_validationRDD: org.apache.spark.rdd.RDD[org.apache.spark.sql.Row] = MapPartitionsRDD[19930] at randomSplit at command-3389902380791621:2
temp_testRDD: org.apache.spark.rdd.RDD[org.apache.spark.sql.Row] = MapPartitionsRDD[19931] at randomSplit at command-3389902380791621:2
// Map the rdds to the Rating type
val maptrainingRDD = temp_trainingRDD.map(x=>Rating(x(0).asInstanceOf[Int], x(1).asInstanceOf[Int], x(2).asInstanceOf[Double]))
val mapvalidationRDD = temp_validationRDD.map(x=>Rating(x(0).asInstanceOf[Int], x(1).asInstanceOf[Int], x(2).asInstanceOf[Double]))
val maptestRDD = temp_testRDD.map(x=>Rating(x(0).asInstanceOf[Int], x(1).asInstanceOf[Int], x(2).asInstanceOf[Double]))
maptrainingRDD: org.apache.spark.rdd.RDD[org.apache.spark.mllib.recommendation.Rating] = MapPartitionsRDD[19932] at map at command-3389902380791624:2
mapvalidationRDD: org.apache.spark.rdd.RDD[org.apache.spark.mllib.recommendation.Rating] = MapPartitionsRDD[19933] at map at command-3389902380791624:3
maptestRDD: org.apache.spark.rdd.RDD[org.apache.spark.mllib.recommendation.Rating] = MapPartitionsRDD[19934] at map at command-3389902380791624:4
// Build the recommendation model using ALS by fitting to the training data (with Hyperparameter tuning)
// trying different hyper-parameter (rank) values to optimise over
val ranks = List(2, 4, 8, 16, 32, 64, 128, 256);
var rank=0;
for ( rank <- ranks ){
// using a fixed numIterations=10 and regularisation=0.01
val numIterations = 10
val regularizationParameter = 0.01
val model = ALS.train(maptrainingRDD, rank, numIterations, regularizationParameter)
// Evaluate the model on test data
val usersProductsValidate = mapvalidationRDD.map { case Rating(user, product, rate) =>
(user, product)
}
// get the predictions on test data
val predictions = model.predict(usersProductsValidate)
.map { case Rating(user, product, rate)
=> ((user, product), rate)
}
// find the actual ratings and join with predictions
val ratesAndPreds = mapvalidationRDD.map { case Rating(user, product, rate)
=> ((user, product), rate)
}.join(predictions)
val MSE = ratesAndPreds.map { case ((user, product), (r1, r2)) =>
val err = (r1 - r2)
err * err
}.mean()
println("rank and Mean Squared Error = " + rank + " and " + MSE)
} // end of loop over ranks
rank and Mean Squared Error = 2 and 1.0624116959469154
rank and Mean Squared Error = 4 and 1.3393495403657538
rank and Mean Squared Error = 8 and 1.6916511125697133
rank and Mean Squared Error = 16 and 1.63542207107039
rank and Mean Squared Error = 32 and 1.311227268934932
rank and Mean Squared Error = 64 and 0.9461947532838
rank and Mean Squared Error = 128 and 0.8859420827613572
rank and Mean Squared Error = 256 and 0.8845268169033572
numIterations: Int = 10
regularisation: Double = 0.01
ranks: List[Int] = List(2, 4, 8, 16, 32, 64, 128, 256)
rank: Int = 0