def analyze(spark, user, items, ratings, faculty):
items_array = items
ratings_array = ratings
# load data
data_util = DataUtil(spark)
db_df = data_util.load_all_df(faculty)
item_df = data_util.get_item_df(db_df)
# db_df.printSchema()
new_df = data_util.create_df_from_new_data(user, items_array, ratings_array, faculty)
# new_df.printSchema()
new_item_df = data_util.get_item_df(new_df)
# print(item_df.count())
input_df = db_df.union(new_df)
# preprocess
input_after_mapping_df = data_util.mapping_course(input_df)
# input_df.show()
item_df = data_util.mapping_course(item_df)
# print(item_df.distinct().count())
new_item_df = data_util.mapping_course(new_item_df)
# data_util.course_mapper.mapping_df.show()
# input_df.printSchema()
# index item
item_indexer = StringIndexer().setInputCol(data_util.get_item_col()).setOutputCol("F_MAMH_index")
item_indexer_model = item_indexer.fit(input_after_mapping_df)
input_index_df = item_indexer_model.transform(input_after_mapping_df) \
.withColumn("F_MAMH_index", col("F_MAMH_index").cast(IntegerType()))
# input_index_df.show()
# input_index_df.printSchema()
# get missing item
missing_item_df = item_indexer_model.transform(item_df.subtract(new_item_df)) \
.withColumn("F_MAMH_index", col("F_MAMH_index").cast(IntegerType())) \
.withColumn("MASV1", lit(user).cast(IntegerType()))
# missing_item_df.show()
# print(item_df.count())
# print(new_item_df.count())
# print(missing_item_df.count())
spark.conf.set("spark.sql.crossJoin.enabled", "true")
# create model
als_nn = ALS(rank=2, maxIter=15, regParam=0.01, userCol="MASV1",
itemCol="F_MAMH_index", ratingCol="TKET", coldStartStrategy="drop", nonnegative=True)
als_model = als_nn.fit(input_index_df)
output_df = als_model.transform(missing_item_df)
# test_df = als_model.transform(input_index_df)
# test_df.show()
# output_df.show()
OutputUtil(spark, "MASV1", "F_MAMH", "prediction").output(output_df, user)
def analyze(spark):
# input_predict_data = [
# (1512400, "CO3059", 10.0),
# (1512400, "CO3031", 9.5),
# (1512400, "CO3055", 9.0),
# (1512400, "CO4027", 9.5),
# (1512400, "CO3029", 8.0),
# (1512400, "CO3021", 10.0),
# (1512400, "IM3001", 9.0),
# (1512400, "MT2001", 7.5),
# (1512400, "SP1007", 8.5),
# (1512400, "MT1005", 8.5),
# (1512400, "PH1003", 7.5),
# (1512400, "CO3043", 0.0),
# (1512400, "CO3025", 1.0),
# (1512400, "CO4313", 2.0)
# ]
# schema = StructType([
# StructField("MASV1", IntegerType(), True),
# StructField("F_MAMH", StringType(), True),
# StructField("TKET", DoubleType(), True)])
# inputDF = spark.createDataFrame(input_predict_data, schema)
data_util = DataUtil(spark)
list_faculty = [
"MT", "BD", "CK", "DC", "DD", "GT", "HC", "MO", "PD", "QL", "UD", "VL",
"VP", "XD"
]
estimator = FPGEstimator(spark, data_util.get_user_col(),
data_util.get_item_col(),
data_util.get_rating_col(), 0.2, 0.8)
for faculty in list_faculty:
data_df = data_util.mapping_course(data_util.load_all_df(faculty))\
.filter(col(data_util.get_rating_col()) >= 5)
transformer = estimator.fit(
data_df.select(data_util.get_user_col(), data_util.get_item_col()))
# transformer.transform(inputDF).show()
transformer.save("model/{}/fp".format(faculty))
def analyze(spark, user, items, ratings, faculty):
data_util = DataUtil(spark)
model_location = "model/{}/nbcf".format(faculty)
item_df = data_util.load_all_df(faculty).select(
data_util.get_item_col()).distinct()
new_data = data_util.mapping_course(
data_util.create_df_from_new_data(user, items, ratings, faculty))
missing_data = item_df.subtract(new_data.select(data_util.get_item_col()).distinct())\
.withColumn(data_util.get_user_col(), lit(user).cast(IntegerType()))
ubcf_model = NBCFTransformer.load(spark, model_location)
output_df = ubcf_model.transform(missing_data)
OutputUtil(spark, "MASV1", "F_MAMH", "prediction").output(output_df, user)
def analyze(spark, user, items, ratings, faculty):
items_array = items
ratings_array = ratings
spark.conf.set("spark.sql.crossJoin.enabled", "true")
model_location = "model/{}/fp".format(faculty)
# load data
data_util = DataUtil(spark)
new_df = data_util.mapping_course(
data_util.create_df_from_new_data(user, items_array, ratings_array,
faculty))
# new_df.show()
model_transformer = FPGTransformer.load(spark, model_location)
recommend_df = model_transformer.transform(new_df)
# recommend_df.show()
# predict rating recommended items
db_df = data_util.mapping_course(data_util.load_all_df(faculty))
# db_df.printSchema()
input_df = db_df.union(new_df).withColumn("TKET",
col("TKET").cast(DoubleType()))
# input_df.printSchema()
# index item
item_indexer = StringIndexer().setInputCol(
data_util.get_item_col()).setOutputCol("F_MAMH_index")
item_indexer_model = item_indexer.fit(input_df)
input_index_df = item_indexer_model.transform(input_df) \
.withColumn("F_MAMH_index", col("F_MAMH_index").cast(IntegerType()))
recommend_df = item_indexer_model.transform(recommend_df)
# create model
als = ALS(rank=2,
maxIter=15,
regParam=0.01,
userCol="MASV1",
itemCol="F_MAMH_index",
ratingCol="TKET",
coldStartStrategy="drop",
nonnegative=True)
als_model = als.fit(input_index_df)
output_df = als_model.transform(recommend_df)
OutputUtil(spark, "MASV1", "F_MAMH", "prediction").output(output_df, user)
def analyze(spark, user, items, ratings, faculty):
data_util = DataUtil(spark)
model_location = "model/{}/als_ibcf".format(faculty)
item_df = data_util.mapping_course(data_util.load_all_df(faculty)).select(
data_util.get_item_col()).distinct()
# start = time.time()
new_data = data_util.mapping_course(
data_util.create_df_from_new_data(user, items, ratings, faculty))
missing_data = item_df.subtract(new_data.select(data_util.get_item_col()).distinct())\
.withColumn(data_util.get_user_col(), lit(user).cast(IntegerType()))
# missing_data.count()
# end = time.time()
# print(str(end - start))
als_ibcf_model = IBCFWithItemFactor(spark)\
.load(model_location)\
.setUserCol(data_util.get_user_col())\
.setItemCol(data_util.get_item_col())\
.setValueCol(data_util.get_rating_col())
output_df = als_ibcf_model.transform(new_data, missing_data)
# print(item_df.
# input_index_df.show()
# input_index_df.printSch
# test_df = als_model.transform(input_index_df)
# test_df.show()
# output_df.show()
OutputUtil(spark, "MASV1", "F_MAMH", "prediction").output(output_df, user)