def _build_ordered_similarity(
self,
select_from,
with_clause_as="_with_clause_normalization_factor"):
"""Build a similarity table col_1, col_2, similarity where col_1 < col_2 """
similarity = SelectQuery()
if self.supports_with_clause:
similarity.with_cte(select_from, alias=with_clause_as)
select_from = with_clause_as
similarity.select_from(select_from,
alias=self.LEFT_NORMALIZATION_FACTOR_AS)
join_conditions = [
Column(self.pivot_column,
self.LEFT_NORMALIZATION_FACTOR_AS).eq_null_unsafe(
Column(self.pivot_column,
self.RIGHT_NORMALIZATION_FACTOR_AS))
]
if self.supports_full_outer_join:
join_conditions += [
Column(self.based_column,
self.LEFT_NORMALIZATION_FACTOR_AS).lt(
Column(self.based_column,
self.RIGHT_NORMALIZATION_FACTOR_AS))
]
else:
join_conditions += [
Column(self.based_column,
self.LEFT_NORMALIZATION_FACTOR_AS).ne(
Column(self.based_column,
self.RIGHT_NORMALIZATION_FACTOR_AS))
]
similarity.join(select_from,
JoinTypes.INNER,
join_conditions,
alias=self.RIGHT_NORMALIZATION_FACTOR_AS)
similarity.group_by(
Column(self.based_column,
table_name=self.LEFT_NORMALIZATION_FACTOR_AS))
similarity.group_by(
Column(self.based_column,
table_name=self.RIGHT_NORMALIZATION_FACTOR_AS))
similarity.select(Column(self.based_column,
table_name=self.LEFT_NORMALIZATION_FACTOR_AS),
alias=f"{self.based_column}_1")
similarity.select(Column(
self.based_column, table_name=self.RIGHT_NORMALIZATION_FACTOR_AS),
alias=f"{self.based_column}_2")
similarity.select(self._get_similarity_formula(),
alias=constants.SIMILARITY_COLUMN_NAME)
return similarity
def make_full_transform_query(aggregation_queries,
dataset,
aggregation_params,
transform_params,
encoding_feature=False):
is_hdfs = 'hiveTableName' in dataset.get_config().get('params').keys()
inner = SelectQuery()
if is_hdfs:
inner.select_from('_'.join(dataset.name.split('.')))
else:
inner.select_from(dataset)
if aggregation_params.is_rolling_window():
inner.select(Column('*'))
else:
inner.distinct() #TODO why?! -> avoid dupplicate
for key in aggregation_params.get_effective_keys():
inner.select(Column(key))
prefilter = _make_prefilter(aggregation_params, transform_params)
inner.where(prefilter)
outer = SelectQuery()
outer.select_from(inner, alias='inner')
if aggregation_params.is_rolling_window():
outer.select(Column('*', 'inner'))
else:
for col in aggregation_params.get_effective_keys(): #+ feature_names:
outer.select(Column(col, 'inner'))
reverse_mapping_dict = {}
for idx, agg_query in enumerate(aggregation_queries):
agg_query.alias(
agg_query.get_alias()
or 'cte_' + str(idx)) #TODO remove, make sure they have ids
outer.with_cte(agg_query)
join_cond = Expression()
for key in aggregation_params.get_effective_keys():
join_cond = join_cond.and_(
Column(key, 'inner').eq_null_unsafe(
Column(key, agg_query.get_alias())))
outer.join(agg_query.get_alias(), JoinTypes.LEFT, join_cond)
for idx2, col in enumerate(agg_query.get_columns_alias()):
if encoding_feature:
if aggregation_params.feature_name_mapping.get(col):
new_alias = '{}_{}_{}'.format(
aggregation_params.feature_name_mapping.get(col), idx,
idx2)
outer.select(Column(col, agg_query.get_alias()), new_alias)
reverse_mapping_dict[new_alias] = col
else:
outer.select(Column(col, agg_query.get_alias()))
return dialectHandler(dataset).convertToSQL(outer), reverse_mapping_dict
def _build_unordered_similarity(
self,
select_from,
left_select_from_as="_left_ordered_similarity",
right_select_from_as="_right_ordered_similarity",
with_clause_as="_with_clause_ordered_similarity",
):
"""Retrieve both pairs (when col_1 < col_2 and col_1 > col_2) from the ordered similarity table"""
similarity = SelectQuery()
if self.supports_with_clause:
similarity.with_cte(select_from, alias=with_clause_as)
select_from = with_clause_as
similarity.select_from(select_from, alias=left_select_from_as)
join_condition = Constant(1).eq_null_unsafe(Constant(0))
similarity.join(select_from,
JoinTypes.FULL,
join_condition,
alias=right_select_from_as)
similarity.select(
Column(f"{self.based_column}_1",
table_name=left_select_from_as).coalesce(
Column(f"{self.based_column}_2",
table_name=right_select_from_as)),
alias=f"{self.based_column}_1",
)
similarity.select(
Column(f"{self.based_column}_2",
table_name=left_select_from_as).coalesce(
Column(f"{self.based_column}_1",
table_name=right_select_from_as)),
alias=f"{self.based_column}_2",
)
similarity.select(
Column("similarity", table_name=left_select_from_as).coalesce(
Column("similarity", table_name=right_select_from_as)),
alias=constants.SIMILARITY_COLUMN_NAME,
)
return similarity
def _build_sum_of_similarity_scores(
self,
top_n,
normalization_factor,
top_n_as="_top_n",
normalization_factor_as="_normalization_factor"):
cf_scores = SelectQuery()
cf_scores.select_from(top_n, alias=top_n_as)
join_condition = Column(f"{self.based_column}_2",
top_n_as).eq_null_unsafe(
Column(self.based_column,
normalization_factor_as))
cf_scores.join(normalization_factor,
JoinTypes.INNER,
join_condition,
alias=normalization_factor_as)
cf_scores.group_by(
Column(f"{self.based_column}_1", table_name=top_n_as))
cf_scores.group_by(
Column(self.pivot_column, table_name=normalization_factor_as))
cf_scores.select(Column(f"{self.based_column}_1", table_name=top_n_as),
alias=self.based_column)
cf_scores.select(
Column(self.pivot_column, table_name=normalization_factor_as))
cf_scores.select(self._get_user_item_similarity_formula(
top_n_as, normalization_factor_as),
alias=constants.SCORE_COLUMN_NAME)
cf_scores.order_by(Column(self.based_column))
cf_scores.order_by(Column(constants.SCORE_COLUMN_NAME),
direction="DESC")
return cf_scores
columns_right = get_recipe_config()['columns_2']
#############################
# Original recipe
#############################
#Start the loop
joins = ['LEFT', 'RIGHT', 'INNER']
join_conds = []
for key in range(len(key_a)):
join_cond = Expression()
globals()['join_cond_'+str(key)] = join_cond.and_(Column(key_a[key], input_A_names[0]).eq_null_unsafe(Column(key_b[key], input_B_names[0])))
join_conds.append(globals()['join_cond_'+str(key)])
for i in joins:
query = SelectQuery()
query.select_from(input_A_datasets[0], alias = input_A_names[0])
for j in columns_left:
query.select(Column(j, input_A_names[0]),alias = j)
for k in columns_right:
query.select(Column(k, input_B_names[0]),alias = k)
query.join(input_B_datasets[0], i, join_conds, operatorBetweenConditions = operator , alias= input_B_names[0])
globals()['sql_'+str(i)] = toSQL(query, input_A_datasets[0])
e = SQLExecutor2()
e.exec_recipe_fragment(output_A_datasets[0], sql_LEFT)
e.exec_recipe_fragment(output_B_datasets[0], sql_RIGHT)
e.exec_recipe_fragment(output_C_datasets[0], sql_INNER)
