背景知识：

1. 协同过滤：简单来说是利用某兴趣相投、拥有共同经验之群体的喜好来推荐用户感兴趣的信息，即群体的智慧
2. 矩阵分解：将（用户、物品、行为）矩阵分解成（用户、隐向量）和（物品，隐向量）两个子矩阵，通过隐向量实现推荐
3. ALS：交替最小二乘法，先假设U的初始值U(0)，可以根据U(0)可以计算出V(0)，再根据V(0)计算出U(1)，迭代到收敛

演示目标：

1. 实现矩阵分解，得到user embedding和item embedding
2. 对于目标user，近邻搜索得到推荐的item列表（需要去除已看、需要查询电影名称）

延伸：

1. user embedding自身的搜索，可以实现兴趣相投的人的推荐
2. item embedding自身的搜索，可以实现相关推荐

import pandas as pd
import numpy as np
import json

import findspark
findspark.init()

from pyspark.sql import SparkSession

spark = SparkSession \
        .builder \
        .appName("Pyspark ALS") \
        .getOrCreate()

sc = spark.sparkContext

from pyspark.sql import functions as F
from pyspark.sql import types as T

customSchema = T.StructType([
    T.StructField("userId",T.IntegerType(),True),
    T.StructField("movieId",T.IntegerType(),True),
    T.StructField("rating",T.FloatType(),True),
    T.StructField("timestamp",T.LongType(),True),
])

df = spark.read.csv(
    "./dataset/datas/ml-latest-small/ratings_1m.csv",
    header =True,
    schema = customSchema
)

df.show(5)

+------+-------+------+---------+
|userId|movieId|rating|timestamp|
+------+-------+------+---------+
|     1|   1193|   5.0|978300760|
|     1|    661|   3.0|978302109|
|     1|    914|   3.0|978301968|
|     1|   3408|   4.0|978300275|
|     1|   2355|   5.0|978824291|
+------+-------+------+---------+
only showing top 5 rows

df.select("userId").distinct().count()

6040

df.select("movieId").distinct().count()

3706

df.printSchema()

root
 |-- userId: integer (nullable = true)
 |-- movieId: integer (nullable = true)
 |-- rating: float (nullable = true)
 |-- timestamp: long (nullable = true)

实现sparkALS的矩阵分解

from pyspark.ml.recommendation import ALS

help(ALS)

Help on class ALS in module pyspark.ml.recommendation:

class ALS(pyspark.ml.wrapper.JavaEstimator, _ALSParams, pyspark.ml.util.JavaMLWritable, pyspark.ml.util.JavaMLReadable)
 |  Alternating Least Squares (ALS) matrix factorization.
 |  
 |  ALS attempts to estimate the ratings matrix `R` as the product of
 |  two lower-rank matrices, `X` and `Y`, i.e. `X * Yt = R`. Typically
 |  these approximations are called 'factor' matrices. The general
 |  approach is iterative. During each iteration, one of the factor
 |  matrices is held constant, while the other is solved for using least
 |  squares. The newly-solved factor matrix is then held constant while
 |  solving for the other factor matrix.
 |  
 |  This is a blocked implementation of the ALS factorization algorithm
 |  that groups the two sets of factors (referred to as "users" and
 |  "products") into blocks and reduces communication by only sending
 |  one copy of each user vector to each product block on each
 |  iteration, and only for the product blocks that need that user's
 |  feature vector. This is achieved by pre-computing some information
 |  about the ratings matrix to determine the "out-links" of each user
 |  (which blocks of products it will contribute to) and "in-link"
 |  information for each product (which of the feature vectors it
 |  receives from each user block it will depend on). This allows us to
 |  send only an array of feature vectors between each user block and
 |  product block, and have the product block find the users' ratings
 |  and update the products based on these messages.
 |  
 |  For implicit preference data, the algorithm used is based on
 |  `"Collaborative Filtering for Implicit Feedback Datasets",
 |  <https://doi.org/10.1109/ICDM.2008.22>`_, adapted for the blocked
 |  approach used here.
 |  
 |  Essentially instead of finding the low-rank approximations to the
 |  rating matrix `R`, this finds the approximations for a preference
 |  matrix `P` where the elements of `P` are 1 if r > 0 and 0 if r <= 0.
 |  The ratings then act as 'confidence' values related to strength of
 |  indicated user preferences rather than explicit ratings given to
 |  items.
 |  
 |  .. versionadded:: 1.4.0
 |  
 |  Notes
 |  -----
 |  The input rating dataframe to the ALS implementation should be deterministic.
 |  Nondeterministic data can cause failure during fitting ALS model.
 |  For example, an order-sensitive operation like sampling after a repartition makes
 |  dataframe output nondeterministic, like `df.repartition(2).sample(False, 0.5, 1618)`.
 |  Checkpointing sampled dataframe or adding a sort before sampling can help make the
 |  dataframe deterministic.
 |  
 |  Examples
 |  --------
 |  >>> df = spark.createDataFrame(
 |  ...     [(0, 0, 4.0), (0, 1, 2.0), (1, 1, 3.0), (1, 2, 4.0), (2, 1, 1.0), (2, 2, 5.0)],
 |  ...     ["user", "item", "rating"])
 |  >>> als = ALS(rank=10, seed=0)
 |  >>> als.setMaxIter(5)
 |  ALS...
 |  >>> als.getMaxIter()
 |  5
 |  >>> als.setRegParam(0.1)
 |  ALS...
 |  >>> als.getRegParam()
 |  0.1
 |  >>> als.clear(als.regParam)
 |  >>> model = als.fit(df)
 |  >>> model.getBlockSize()
 |  4096
 |  >>> model.getUserCol()
 |  'user'
 |  >>> model.setUserCol("user")
 |  ALSModel...
 |  >>> model.getItemCol()
 |  'item'
 |  >>> model.setPredictionCol("newPrediction")
 |  ALS...
 |  >>> model.rank
 |  10
 |  >>> model.userFactors.orderBy("id").collect()
 |  [Row(id=0, features=[...]), Row(id=1, ...), Row(id=2, ...)]
 |  >>> test = spark.createDataFrame([(0, 2), (1, 0), (2, 0)], ["user", "item"])
 |  >>> predictions = sorted(model.transform(test).collect(), key=lambda r: r[0])
 |  >>> predictions[0]
 |  Row(user=0, item=2, newPrediction=0.692910...)
 |  >>> predictions[1]
 |  Row(user=1, item=0, newPrediction=3.473569...)
 |  >>> predictions[2]
 |  Row(user=2, item=0, newPrediction=-0.899198...)
 |  >>> user_recs = model.recommendForAllUsers(3)
 |  >>> user_recs.where(user_recs.user == 0)        .select("recommendations.item", "recommendations.rating").collect()
 |  [Row(item=[0, 1, 2], rating=[3.910..., 1.997..., 0.692...])]
 |  >>> item_recs = model.recommendForAllItems(3)
 |  >>> item_recs.where(item_recs.item == 2)        .select("recommendations.user", "recommendations.rating").collect()
 |  [Row(user=[2, 1, 0], rating=[4.892..., 3.991..., 0.692...])]
 |  >>> user_subset = df.where(df.user == 2)
 |  >>> user_subset_recs = model.recommendForUserSubset(user_subset, 3)
 |  >>> user_subset_recs.select("recommendations.item", "recommendations.rating").first()
 |  Row(item=[2, 1, 0], rating=[4.892..., 1.076..., -0.899...])
 |  >>> item_subset = df.where(df.item == 0)
 |  >>> item_subset_recs = model.recommendForItemSubset(item_subset, 3)
 |  >>> item_subset_recs.select("recommendations.user", "recommendations.rating").first()
 |  Row(user=[0, 1, 2], rating=[3.910..., 3.473..., -0.899...])
 |  >>> als_path = temp_path + "/als"
 |  >>> als.save(als_path)
 |  >>> als2 = ALS.load(als_path)
 |  >>> als.getMaxIter()
 |  5
 |  >>> model_path = temp_path + "/als_model"
 |  >>> model.save(model_path)
 |  >>> model2 = ALSModel.load(model_path)
 |  >>> model.rank == model2.rank
 |  True
 |  >>> sorted(model.userFactors.collect()) == sorted(model2.userFactors.collect())
 |  True
 |  >>> sorted(model.itemFactors.collect()) == sorted(model2.itemFactors.collect())
 |  True
 |  >>> model.transform(test).take(1) == model2.transform(test).take(1)
 |  True
 |  
 |  Method resolution order:
 |      ALS
 |      pyspark.ml.wrapper.JavaEstimator
 |      pyspark.ml.wrapper.JavaParams
 |      pyspark.ml.wrapper.JavaWrapper
 |      pyspark.ml.base.Estimator
 |      _ALSParams
 |      _ALSModelParams
 |      pyspark.ml.param.shared.HasPredictionCol
 |      pyspark.ml.param.shared.HasBlockSize
 |      pyspark.ml.param.shared.HasMaxIter
 |      pyspark.ml.param.shared.HasRegParam
 |      pyspark.ml.param.shared.HasCheckpointInterval
 |      pyspark.ml.param.shared.HasSeed
 |      pyspark.ml.param.Params
 |      pyspark.ml.util.Identifiable
 |      pyspark.ml.util.JavaMLWritable
 |      pyspark.ml.util.MLWritable
 |      pyspark.ml.util.JavaMLReadable
 |      pyspark.ml.util.MLReadable
 |      builtins.object
 |  
 |  Methods defined here:
 |  
 |  __init__(self, *, rank=10, maxIter=10, regParam=0.1, numUserBlocks=10, numItemBlocks=10, implicitPrefs=False, alpha=1.0, userCol='user', itemCol='item', seed=None, ratingCol='rating', nonnegative=False, checkpointInterval=10, intermediateStorageLevel='MEMORY_AND_DISK', finalStorageLevel='MEMORY_AND_DISK', coldStartStrategy='nan', blockSize=4096)
 |      __init__(self, \*, rank=10, maxIter=10, regParam=0.1, numUserBlocks=10,
 |               numItemBlocks=10, implicitPrefs=False, alpha=1.0, userCol="user", itemCol="item",                  seed=None, ratingCol="rating", nonnegative=False, checkpointInterval=10,                  intermediateStorageLevel="MEMORY_AND_DISK",                  finalStorageLevel="MEMORY_AND_DISK", coldStartStrategy="nan", blockSize=4096)
 |  
 |  setAlpha(self, value)
 |      Sets the value of :py:attr:`alpha`.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setBlockSize(self, value)
 |      Sets the value of :py:attr:`blockSize`.
 |      
 |      .. versionadded:: 3.0.0
 |  
 |  setCheckpointInterval(self, value)
 |      Sets the value of :py:attr:`checkpointInterval`.
 |  
 |  setColdStartStrategy(self, value)
 |      Sets the value of :py:attr:`coldStartStrategy`.
 |      
 |      .. versionadded:: 2.2.0
 |  
 |  setFinalStorageLevel(self, value)
 |      Sets the value of :py:attr:`finalStorageLevel`.
 |      
 |      .. versionadded:: 2.0.0
 |  
 |  setImplicitPrefs(self, value)
 |      Sets the value of :py:attr:`implicitPrefs`.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setIntermediateStorageLevel(self, value)
 |      Sets the value of :py:attr:`intermediateStorageLevel`.
 |      
 |      .. versionadded:: 2.0.0
 |  
 |  setItemCol(self, value)
 |      Sets the value of :py:attr:`itemCol`.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setMaxIter(self, value)
 |      Sets the value of :py:attr:`maxIter`.
 |  
 |  setNonnegative(self, value)
 |      Sets the value of :py:attr:`nonnegative`.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setNumBlocks(self, value)
 |      Sets both :py:attr:`numUserBlocks` and :py:attr:`numItemBlocks` to the specific value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setNumItemBlocks(self, value)
 |      Sets the value of :py:attr:`numItemBlocks`.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setNumUserBlocks(self, value)
 |      Sets the value of :py:attr:`numUserBlocks`.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setParams(self, *, rank=10, maxIter=10, regParam=0.1, numUserBlocks=10, numItemBlocks=10, implicitPrefs=False, alpha=1.0, userCol='user', itemCol='item', seed=None, ratingCol='rating', nonnegative=False, checkpointInterval=10, intermediateStorageLevel='MEMORY_AND_DISK', finalStorageLevel='MEMORY_AND_DISK', coldStartStrategy='nan', blockSize=4096)
 |      setParams(self, \*, rank=10, maxIter=10, regParam=0.1, numUserBlocks=10,                  numItemBlocks=10, implicitPrefs=False, alpha=1.0, userCol="user", itemCol="item",                  seed=None, ratingCol="rating", nonnegative=False, checkpointInterval=10,                  intermediateStorageLevel="MEMORY_AND_DISK",                  finalStorageLevel="MEMORY_AND_DISK", coldStartStrategy="nan", blockSize=4096)
 |      Sets params for ALS.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setPredictionCol(self, value)
 |      Sets the value of :py:attr:`predictionCol`.
 |  
 |  setRank(self, value)
 |      Sets the value of :py:attr:`rank`.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setRatingCol(self, value)
 |      Sets the value of :py:attr:`ratingCol`.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  setRegParam(self, value)
 |      Sets the value of :py:attr:`regParam`.
 |  
 |  setSeed(self, value)
 |      Sets the value of :py:attr:`seed`.
 |  
 |  setUserCol(self, value)
 |      Sets the value of :py:attr:`userCol`.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  ----------------------------------------------------------------------
 |  Data and other attributes defined here:
 |  
 |  __abstractmethods__ = frozenset()
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.wrapper.JavaParams:
 |  
 |  clear(self, param)
 |      Clears a param from the param map if it has been explicitly set.
 |  
 |  copy(self, extra=None)
 |      Creates a copy of this instance with the same uid and some
 |      extra params. This implementation first calls Params.copy and
 |      then make a copy of the companion Java pipeline component with
 |      extra params. So both the Python wrapper and the Java pipeline
 |      component get copied.
 |      
 |      Parameters
 |      ----------
 |      extra : dict, optional
 |          Extra parameters to copy to the new instance
 |      
 |      Returns
 |      -------
 |      :py:class:`JavaParams`
 |          Copy of this instance
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.wrapper.JavaWrapper:
 |  
 |  __del__(self)
 |  
 |  ----------------------------------------------------------------------
 |  Data descriptors inherited from pyspark.ml.wrapper.JavaWrapper:
 |  
 |  __dict__
 |      dictionary for instance variables (if defined)
 |  
 |  __weakref__
 |      list of weak references to the object (if defined)
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.base.Estimator:
 |  
 |  fit(self, dataset, params=None)
 |      Fits a model to the input dataset with optional parameters.
 |      
 |      .. versionadded:: 1.3.0
 |      
 |      Parameters
 |      ----------
 |      dataset : :py:class:`pyspark.sql.DataFrame`
 |          input dataset.
 |      params : dict or list or tuple, optional
 |          an optional param map that overrides embedded params. If a list/tuple of
 |          param maps is given, this calls fit on each param map and returns a list of
 |          models.
 |      
 |      Returns
 |      -------
 |      :py:class:`Transformer` or a list of :py:class:`Transformer`
 |          fitted model(s)
 |  
 |  fitMultiple(self, dataset, paramMaps)
 |      Fits a model to the input dataset for each param map in `paramMaps`.
 |      
 |      .. versionadded:: 2.3.0
 |      
 |      Parameters
 |      ----------
 |      dataset : :py:class:`pyspark.sql.DataFrame`
 |          input dataset.
 |      paramMaps : :py:class:`collections.abc.Sequence`
 |          A Sequence of param maps.
 |      
 |      Returns
 |      -------
 |      :py:class:`_FitMultipleIterator`
 |          A thread safe iterable which contains one model for each param map. Each
 |          call to `next(modelIterator)` will return `(index, model)` where model was fit
 |          using `paramMaps[index]`. `index` values may not be sequential.
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from _ALSParams:
 |  
 |  getAlpha(self)
 |      Gets the value of alpha or its default value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  getFinalStorageLevel(self)
 |      Gets the value of finalStorageLevel or its default value.
 |      
 |      .. versionadded:: 2.0.0
 |  
 |  getImplicitPrefs(self)
 |      Gets the value of implicitPrefs or its default value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  getIntermediateStorageLevel(self)
 |      Gets the value of intermediateStorageLevel or its default value.
 |      
 |      .. versionadded:: 2.0.0
 |  
 |  getNonnegative(self)
 |      Gets the value of nonnegative or its default value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  getNumItemBlocks(self)
 |      Gets the value of numItemBlocks or its default value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  getNumUserBlocks(self)
 |      Gets the value of numUserBlocks or its default value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  getRank(self)
 |      Gets the value of rank or its default value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  getRatingCol(self)
 |      Gets the value of ratingCol or its default value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  ----------------------------------------------------------------------
 |  Data and other attributes inherited from _ALSParams:
 |  
 |  alpha = Param(parent='undefined', name='alpha', doc='alpha for implici...
 |  
 |  finalStorageLevel = Param(parent='undefined', name='finalStorageLevel'...
 |  
 |  implicitPrefs = Param(parent='undefined', name='implicitPrefs', doc='w...
 |  
 |  intermediateStorageLevel = Param(parent='undefined', name='intermediat...
 |  
 |  nonnegative = Param(parent='undefined', name='nonnegative', do...to us...
 |  
 |  numItemBlocks = Param(parent='undefined', name='numItemBlocks', doc='n...
 |  
 |  numUserBlocks = Param(parent='undefined', name='numUserBlocks', doc='n...
 |  
 |  rank = Param(parent='undefined', name='rank', doc='rank of the factori...
 |  
 |  ratingCol = Param(parent='undefined', name='ratingCol', doc='column na...
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from _ALSModelParams:
 |  
 |  getColdStartStrategy(self)
 |      Gets the value of coldStartStrategy or its default value.
 |      
 |      .. versionadded:: 2.2.0
 |  
 |  getItemCol(self)
 |      Gets the value of itemCol or its default value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  getUserCol(self)
 |      Gets the value of userCol or its default value.
 |      
 |      .. versionadded:: 1.4.0
 |  
 |  ----------------------------------------------------------------------
 |  Data and other attributes inherited from _ALSModelParams:
 |  
 |  coldStartStrategy = Param(parent='undefined', name='coldStartStrateg.....
 |  
 |  itemCol = Param(parent='undefined', name='itemCol', doc='c...ds. Ids m...
 |  
 |  userCol = Param(parent='undefined', name='userCol', doc='c...ds. Ids m...
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.param.shared.HasPredictionCol:
 |  
 |  getPredictionCol(self)
 |      Gets the value of predictionCol or its default value.
 |  
 |  ----------------------------------------------------------------------
 |  Data and other attributes inherited from pyspark.ml.param.shared.HasPredictionCol:
 |  
 |  predictionCol = Param(parent='undefined', name='predictionCol', doc='p...
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.param.shared.HasBlockSize:
 |  
 |  getBlockSize(self)
 |      Gets the value of blockSize or its default value.
 |  
 |  ----------------------------------------------------------------------
 |  Data and other attributes inherited from pyspark.ml.param.shared.HasBlockSize:
 |  
 |  blockSize = Param(parent='undefined', name='blockSize', doc=...n then ...
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.param.shared.HasMaxIter:
 |  
 |  getMaxIter(self)
 |      Gets the value of maxIter or its default value.
 |  
 |  ----------------------------------------------------------------------
 |  Data and other attributes inherited from pyspark.ml.param.shared.HasMaxIter:
 |  
 |  maxIter = Param(parent='undefined', name='maxIter', doc='max number of...
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.param.shared.HasRegParam:
 |  
 |  getRegParam(self)
 |      Gets the value of regParam or its default value.
 |  
 |  ----------------------------------------------------------------------
 |  Data and other attributes inherited from pyspark.ml.param.shared.HasRegParam:
 |  
 |  regParam = Param(parent='undefined', name='regParam', doc='regularizat...
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.param.shared.HasCheckpointInterval:
 |  
 |  getCheckpointInterval(self)
 |      Gets the value of checkpointInterval or its default value.
 |  
 |  ----------------------------------------------------------------------
 |  Data and other attributes inherited from pyspark.ml.param.shared.HasCheckpointInterval:
 |  
 |  checkpointInterval = Param(parent='undefined', name='checkpointInterv....
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.param.shared.HasSeed:
 |  
 |  getSeed(self)
 |      Gets the value of seed or its default value.
 |  
 |  ----------------------------------------------------------------------
 |  Data and other attributes inherited from pyspark.ml.param.shared.HasSeed:
 |  
 |  seed = Param(parent='undefined', name='seed', doc='random seed.')
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.param.Params:
 |  
 |  explainParam(self, param)
 |      Explains a single param and returns its name, doc, and optional
 |      default value and user-supplied value in a string.
 |  
 |  explainParams(self)
 |      Returns the documentation of all params with their optionally
 |      default values and user-supplied values.
 |  
 |  extractParamMap(self, extra=None)
 |      Extracts the embedded default param values and user-supplied
 |      values, and then merges them with extra values from input into
 |      a flat param map, where the latter value is used if there exist
 |      conflicts, i.e., with ordering: default param values <
 |      user-supplied values < extra.
 |      
 |      Parameters
 |      ----------
 |      extra : dict, optional
 |          extra param values
 |      
 |      Returns
 |      -------
 |      dict
 |          merged param map
 |  
 |  getOrDefault(self, param)
 |      Gets the value of a param in the user-supplied param map or its
 |      default value. Raises an error if neither is set.
 |  
 |  getParam(self, paramName)
 |      Gets a param by its name.
 |  
 |  hasDefault(self, param)
 |      Checks whether a param has a default value.
 |  
 |  hasParam(self, paramName)
 |      Tests whether this instance contains a param with a given
 |      (string) name.
 |  
 |  isDefined(self, param)
 |      Checks whether a param is explicitly set by user or has
 |      a default value.
 |  
 |  isSet(self, param)
 |      Checks whether a param is explicitly set by user.
 |  
 |  set(self, param, value)
 |      Sets a parameter in the embedded param map.
 |  
 |  ----------------------------------------------------------------------
 |  Data descriptors inherited from pyspark.ml.param.Params:
 |  
 |  params
 |      Returns all params ordered by name. The default implementation
 |      uses :py:func:`dir` to get all attributes of type
 |      :py:class:`Param`.
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.util.Identifiable:
 |  
 |  __repr__(self)
 |      Return repr(self).
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.util.JavaMLWritable:
 |  
 |  write(self)
 |      Returns an MLWriter instance for this ML instance.
 |  
 |  ----------------------------------------------------------------------
 |  Methods inherited from pyspark.ml.util.MLWritable:
 |  
 |  save(self, path)
 |      Save this ML instance to the given path, a shortcut of 'write().save(path)'.
 |  
 |  ----------------------------------------------------------------------
 |  Class methods inherited from pyspark.ml.util.JavaMLReadable:
 |  
 |  read() from abc.ABCMeta
 |      Returns an MLReader instance for this class.
 |  
 |  ----------------------------------------------------------------------
 |  Class methods inherited from pyspark.ml.util.MLReadable:
 |  
 |  load(path) from abc.ABCMeta
 |      Reads an ML instance from the input path, a shortcut of `read().load(path)`.

als = ALS(
    regParam=0.01,
    userCol = "userId",
    itemCol = "movieId",
    ratingCol = "rating",
    coldStartStrategy="drop"
)

model = als.fit(df)

model

ALSModel: uid=ALS_e8292c58bc5f, rank=10

model.userFactors.show(5)

+---+--------------------+
| id|            features|
+---+--------------------+
| 10|[0.21681188, 1.20...|
| 20|[0.20809321, 0.89...|
| 30|[0.16379297, 0.26...|
| 40|[0.007984038, 1.2...|
| 50|[1.0053813, 0.767...|
+---+--------------------+
only showing top 5 rows

model.userFactors.count()

6040

model.itemFactors.count()

3706

model.userFactors.select("id","features") \
        .toPandas() \
        .to_csv("data_sparkals_user_embedding.csv",index=False)

model.itemFactors.select("id","features") \
        .toPandas() \
        .to_csv("data_sparkals_item_embedding.csv",index=False)

对于给定用户算出可能最喜欢的10个电影

思路：

1. 查询目标用户的embedding
2. 计算目标用户embedding跟所有movie embedding的sim value
3. 计算用户看过的集合
4. 第2步骤过滤掉看过的集合，然后挑选出前10个电影

target_user_id = 1

df_movie = pd.read_csv("./dataset/datas/ml-latest-small/movies.csv")
df_rating = pd.read_csv("./dataset/datas/ml-latest-small/ratings.csv")

df_movie_embedding = pd.read_csv("./data_sparkals_item_embedding.csv")
df_user_embedding = pd.read_csv("./data_sparkals_user_embedding.csv")

df_movie_embedding.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 3706 entries, 0 to 3705
Data columns (total 2 columns):
 #   Column    Non-Null Count  Dtype 
---  ------    --------------  ----- 
 0   id        3706 non-null   int64 
 1   features  3706 non-null   object
dtypes: int64(1), object(1)
memory usage: 58.0+ KB

df_movie_embedding

	id	features
0	10	[-0.18815411627292633, 0.3980417847633362, 0.4...
1	20	[-0.2916702330112457, 0.39180925488471985, -0....
2	30	[-0.15932556986808777, 1.0650469064712524, -0....
3	40	[0.5688132047653198, 0.6850257515907288, 0.328...
4	50	[0.26137569546699524, 0.10741746425628662, 0.2...
...	...	...
3701	3909	[-0.37682488560676575, 0.9805638790130615, -0....
3702	3919	[-0.2180541306734085, 0.9555624723434448, 0.72...
3703	3929	[0.2916938066482544, 1.0744397640228271, 0.254...
3704	3939	[-0.7031605243682861, 1.002614140510559, 0.204...
3705	3949	[0.4849887490272522, 0.5978550314903259, -0.02...

3706 rows × 2 columns

df_user_embedding

	id	features
0	10	[0.2168118804693222, 1.2006480693817139, 0.790...
1	20	[0.20809321105480194, 0.8956680297851562, 1.48...
2	30	[0.16379296779632568, 0.26286977529525757, 0.7...
3	40	[0.007984038442373276, 1.251495122909546, 0.88...
4	50	[1.0053813457489014, 0.7675292491912842, 0.526...
...	...	...
6035	5999	[0.5730876922607422, 1.1653149127960205, 0.781...
6036	6009	[0.17401114106178284, 0.4095298945903778, 0.16...
6037	6019	[-0.5992759466171265, 1.0960370302200317, 0.21...
6038	6029	[1.2999039888381958, 0.9374491572380066, 0.136...
6039	6039	[0.2989945411682129, 0.741441547870636, 0.8007...

6040 rows × 2 columns

# embeding 从字符串向量化
df_movie_embedding["features"] = df_movie_embedding["features"].map(lambda x : np.array(json.loads(x)))
df_user_embedding["features"] = df_user_embedding["features"].map(lambda x : np.array(json.loads(x)))

df_user_embedding.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 6040 entries, 0 to 6039
Data columns (total 2 columns):
 #   Column    Non-Null Count  Dtype 
---  ------    --------------  ----- 
 0   id        6040 non-null   int64 
 1   features  6040 non-null   object
dtypes: int64(1), object(1)
memory usage: 94.5+ KB

df_user_embedding[df_user_embedding["id"]==target_user_id]["features"].iloc[0]

array([-0.68436241,  0.87026972, -0.37516031, -1.56462491,  0.46863654,
       -0.95500118,  1.14803946, -0.41115561, -2.45734978,  0.70034939])

计算 user embedding 和 item embedding 的相似度

user_embedding=df_user_embedding[df_user_embedding["id"]==1]["features"].iloc[0]

from scipy.spatial import distance

df_movie_embedding["sim_value"] =df_movie_embedding["features"].map(lambda x : 1-distance.cosine(user_embedding,x))

df_movie_embedding

	id	features	sim_value
0	10	[-0.18815411627292633, 0.3980417847633362, 0.4...	0.705337
1	20	[-0.2916702330112457, 0.39180925488471985, -0....	0.708213
2	30	[-0.15932556986808777, 1.0650469064712524, -0....	0.629950
3	40	[0.5688132047653198, 0.6850257515907288, 0.328...	0.408502
4	50	[0.26137569546699524, 0.10741746425628662, 0.2...	0.669305
...	...	...	...
3701	3909	[-0.37682488560676575, 0.9805638790130615, -0....	0.631404
3702	3919	[-0.2180541306734085, 0.9555624723434448, 0.72...	0.290478
3703	3929	[0.2916938066482544, 1.0744397640228271, 0.254...	0.487208
3704	3939	[-0.7031605243682861, 1.002614140510559, 0.204...	0.367001
3705	3949	[0.4849887490272522, 0.5978550314903259, -0.02...	0.487794

3706 rows × 3 columns

df_rating[df_rating["userId"] == target_user_id]["movieId"].unique()

array([   1,    3,    6,   47,   50,   70,  101,  110,  151,  157,  163,
        216,  223,  231,  235,  260,  296,  316,  333,  349,  356,  362,
        367,  423,  441,  457,  480,  500,  527,  543,  552,  553,  590,
        592,  593,  596,  608,  648,  661,  673,  733,  736,  780,  804,
        919,  923,  940,  943,  954, 1009, 1023, 1024, 1025, 1029, 1030,
       1031, 1032, 1042, 1049, 1060, 1073, 1080, 1089, 1090, 1092, 1097,
       1127, 1136, 1196, 1197, 1198, 1206, 1208, 1210, 1213, 1214, 1219,
       1220, 1222, 1224, 1226, 1240, 1256, 1258, 1265, 1270, 1275, 1278,
       1282, 1291, 1298, 1348, 1377, 1396, 1408, 1445, 1473, 1500, 1517,
       1552, 1573, 1580, 1587, 1617, 1620, 1625, 1644, 1676, 1732, 1777,
       1793, 1804, 1805, 1920, 1927, 1954, 1967, 2000, 2005, 2012, 2018,
       2028, 2033, 2046, 2048, 2054, 2058, 2078, 2090, 2093, 2094, 2096,
       2099, 2105, 2115, 2116, 2137, 2139, 2141, 2143, 2161, 2174, 2193,
       2253, 2268, 2273, 2291, 2329, 2338, 2353, 2366, 2387, 2389, 2395,
       2406, 2414, 2427, 2450, 2459, 2470, 2478, 2492, 2502, 2528, 2529,
       2542, 2571, 2580, 2596, 2616, 2617, 2628, 2640, 2641, 2644, 2648,
       2654, 2657, 2692, 2700, 2716, 2761, 2797, 2826, 2858, 2872, 2899,
       2916, 2944, 2947, 2948, 2949, 2959, 2985, 2987, 2991, 2993, 2997,
       3033, 3034, 3052, 3053, 3062, 3147, 3168, 3176, 3243, 3247, 3253,
       3273, 3386, 3439, 3440, 3441, 3448, 3450, 3479, 3489, 3527, 3578,
       3617, 3639, 3671, 3702, 3703, 3729, 3740, 3744, 3793, 3809, 4006,
       5060], dtype=int64)

watched_ids = set(df_rating[df_rating["userId"] == target_user_id]["movieId"].unique())

len(watched_ids)

232

df_movie_embedding[~df_movie_embedding["id"].isin(watched_ids)]

	id	features	sim_value
0	10	[-0.18815411627292633, 0.3980417847633362, 0.4...	0.705337
1	20	[-0.2916702330112457, 0.39180925488471985, -0....	0.708213
2	30	[-0.15932556986808777, 1.0650469064712524, -0....	0.629950
3	40	[0.5688132047653198, 0.6850257515907288, 0.328...	0.408502
5	60	[0.2520754635334015, 0.7954497337341309, -0.42...	0.817684
...	...	...	...
3701	3909	[-0.37682488560676575, 0.9805638790130615, -0....	0.631404
3702	3919	[-0.2180541306734085, 0.9555624723434448, 0.72...	0.290478
3703	3929	[0.2916938066482544, 1.0744397640228271, 0.254...	0.487208
3704	3939	[-0.7031605243682861, 1.002614140510559, 0.204...	0.367001
3705	3949	[0.4849887490272522, 0.5978550314903259, -0.02...	0.487794

3476 rows × 3 columns

df_movie_embedding[df_movie_embedding["id"].isin(watched_ids)]

	id	features	sim_value
4	50	[0.26137569546699524, 0.10741746425628662, 0.2...	0.669305
6	70	[-0.3985518217086792, 0.8255515098571777, 1.23...	0.335697
10	110	[-0.3063579797744751, -0.16125208139419556, 0....	0.789138
25	260	[0.05825613811612129, 0.012431434355676174, 0....	0.617702
46	480	[0.115965835750103, 0.12412750720977783, 0.765...	0.710630
...	...	...	...
3660	3479	[0.1421741545200348, 0.4269544184207916, 0.144...	0.692919
3661	3489	[-0.2083592563867569, 0.229205921292305, 0.163...	0.613882
3675	3639	[-0.3138227164745331, 0.09211219847202301, 0.1...	0.707233
3684	3729	[-0.14264270663261414, 0.3568372130393982, -0....	0.857207
3692	3809	[-0.23935143649578094, 0.6940127611160278, -0....	0.715511

230 rows × 3 columns

df_movie_embedding[~df_movie_embedding["id"].isin(watched_ids)].sort_values("sim_value",ascending=False).head(10)

	id	features	sim_value
753	122	[-0.2713378965854645, 0.5307934284210205, -0.3...	0.952694
1298	2013	[-0.22150848805904388, 0.17947401106357574, -0...	0.946765
299	3250	[0.10610243678092957, 0.289923757314682, -0.08...	0.933755
221	2420	[-0.11612560600042343, 0.21996952593326569, 0....	0.932687
3384	479	[-0.44652122259140015, 0.4711386561393738, -0....	0.932301
1687	2114	[-0.22058504819869995, 0.32801997661590576, -0...	0.931865
138	1470	[-0.18237100541591644, 0.39677849411964417, 0....	0.927258
325	3510	[-0.18631470203399658, 0.19551320374011993, 0....	0.918885
3187	2358	[-0.5845286250114441, 0.6346001625061035, 0.02...	0.917102
1339	2423	[-0.624200701713562, 0.22440963983535767, 0.12...	0.910244

df_target_movieIds = df_movie_embedding[~df_movie_embedding["id"].isin(watched_ids)].sort_values("sim_value",ascending=False).head(10)[["id","sim_value"]]

df_target_movieIds

	id	sim_value
753	122	0.952694
1298	2013	0.946765
299	3250	0.933755
221	2420	0.932687
3384	479	0.932301
1687	2114	0.931865
138	1470	0.927258
325	3510	0.918885
3187	2358	0.917102
1339	2423	0.910244

查询ID的电影名和信息展现给用户

df_target_movieIds

	id	sim_value
753	122	0.952694
1298	2013	0.946765
299	3250	0.933755
221	2420	0.932687
3384	479	0.932301
1687	2114	0.931865
138	1470	0.927258
325	3510	0.918885
3187	2358	0.917102
1339	2423	0.910244

pd.merge(left=df_target_movieIds,right=df_movie,left_on="id",right_on="movieId")[["movieId","title","genres","sim_value"]]

	movieId	title	genres	sim_value
0	122	Boomerang (1992)	Comedy|Romance	0.952694
1	2013	Poseidon Adventure, The (1972)	Action|Adventure|Drama	0.946765
2	3250	Alive (1993)	Drama	0.933755
3	2420	Karate Kid, The (1984)	Drama	0.932687
4	479	Judgment Night (1993)	Action|Crime|Thriller	0.932301
5	2114	Outsiders, The (1983)	Drama	0.931865
6	3510	Frequency (2000)	Drama|Thriller	0.918885
7	2358	Savior (1998)	Drama|War	0.917102
8	2423	Christmas Vacation (National Lampoon's Christm...	Comedy	0.910244

df_rating[df_rating["userId"]==target_user_id]

	userId	movieId	rating	timestamp
0	1	1	4.0	964982703
1	1	3	4.0	964981247
2	1	6	4.0	964982224
3	1	47	5.0	964983815
4	1	50	5.0	964982931
...	...	...	...	...
227	1	3744	4.0	964980694
228	1	3793	5.0	964981855
229	1	3809	4.0	964981220
230	1	4006	4.0	964982903
231	1	5060	5.0	964984002

232 rows × 4 columns

df_movie[df_movie["movieId"].isin(watched_ids)].merge(df_rating[df_rating["userId"]==target_user_id], on="movieId")

	movieId	title	genres	userId	rating	timestamp
0	1	Toy Story (1995)	Adventure|Animation|Children|Comedy|Fantasy	1	4.0	964982703
1	3	Grumpier Old Men (1995)	Comedy|Romance	1	4.0	964981247
2	6	Heat (1995)	Action|Crime|Thriller	1	4.0	964982224
3	47	Seven (a.k.a. Se7en) (1995)	Mystery|Thriller	1	5.0	964983815
4	50	Usual Suspects, The (1995)	Crime|Mystery|Thriller	1	5.0	964982931
...	...	...	...	...	...	...
227	3744	Shaft (2000)	Action|Crime|Thriller	1	4.0	964980694
228	3793	X-Men (2000)	Action|Adventure|Sci-Fi	1	5.0	964981855
229	3809	What About Bob? (1991)	Comedy	1	4.0	964981220
230	4006	Transformers: The Movie (1986)	Adventure|Animation|Children|Sci-Fi	1	4.0	964982903
231	5060	M*A*S*H (a.k.a. MASH) (1970)	Comedy|Drama|War	1	5.0	964984002

232 rows × 6 columns

df_movie[df_movie["movieId"].isin(watched_ids)].merge(df_rating[df_rating["userId"]==target_user_id], on="movieId").sort_values("rating",ascending=False)

	movieId	title	genres	userId	rating	timestamp
231	5060	M*A*S*H (a.k.a. MASH) (1970)	Comedy|Drama|War	1	5.0	964984002
185	2872	Excalibur (1981)	Adventure|Fantasy	1	5.0	964981680
89	1291	Indiana Jones and the Last Crusade (1989)	Action|Adventure	1	5.0	964981909
90	1298	Pink Floyd: The Wall (1982)	Drama|Musical	1	5.0	964984086
190	2948	From Russia with Love (1963)	Action|Adventure|Thriller	1	5.0	964982191
...	...	...	...	...	...	...
170	2617	Mummy, The (1999)	Action|Adventure|Comedy|Fantasy|Horror|Thriller	1	2.0	964982588
143	2253	Toys (1992)	Comedy|Fantasy	1	2.0	964981775
148	2338	I Still Know What You Did Last Summer (1998)	Horror|Mystery|Thriller	1	2.0	964983546
152	2389	Psycho (1998)	Crime|Horror|Thriller	1	2.0	964983094
205	3176	Talented Mr. Ripley, The (1999)	Drama|Mystery|Thriller	1	1.0	964983504

232 rows × 6 columns