def entries(self, conditions="1=1", order_by=None, limit=None, offset=0, step=10000, fields=None): """ Generate a fully denormalized view of the entries on this table. This view is nested so that each dimension will be a hash of its attributes. This is somewhat similar to the entries collection in the fully denormalized schema before OpenSpending 0.11 (MongoDB). """ if not self.is_generated: return if fields is None: fields = self.fields joins = self.alias for d in self.dimensions: if d in fields: joins = d.join(joins) selects = [f.selectable for f in fields] + [self.alias.c.id] # enforce stable sorting: if order_by is None: order_by = [self.alias.c.id.asc()] for i in count(): qoffset = offset + (step * i) qlimit = step if limit is not None: qlimit = min(limit - (step * i), step) if qlimit <= 0: break query = db.select(selects, conditions, joins, order_by=order_by, use_labels=True, limit=qlimit, offset=qoffset) rp = self.bind.execute(query) first_row = True while True: row = rp.fetchone() if row is None: if first_row: return break first_row = False yield decode_row(row, self)
def entries(self, conditions="1=1", order_by=None, limit=None, offset=0, step=10000, fields=None): """ Generate a fully denormalized view of the entries on this table. This view is nested so that each dimension will be a hash of its attributes. This is somewhat similar to the entries collection in the fully denormalized schema before OpenSpending 0.11 (MongoDB). """ if not self.is_generated: return if fields is None: fields = self.fields joins = self.alias for d in self.dimensions: if d in fields: joins = d.join(joins) selects = [f.selectable for f in fields] + [self.alias.c.id] # enforce stable sorting: if order_by is None: order_by = [self.alias.c.id.asc()] for i in count(): qoffset = offset + (step * i) qlimit = step if limit is not None: qlimit = min(limit - (step * i), step) if qlimit <= 0: break query = select(selects, conditions, joins, order_by=order_by, use_labels=True, limit=qlimit, offset=qoffset) rp = self.bind.execute(query) first_row = True while True: row = rp.fetchone() if row is None: if first_row: return break first_row = False yield decode_row(row, self)
def aggregate(self, measures=['amount'], drilldowns=[], cuts=[], page=1, pagesize=10000, order=[]): """ Query the dataset for a subset of cells based on cuts and drilldowns. It returns a structure with a list of drilldown items and a summary about the slice cutted by the query. ``measures`` The numeric units to be aggregated over, defaults to [``amount``]. (type: `list`) ``drilldowns`` Dimensions to drill down to. (type: `list`) ``cuts`` Specification what to cut from the cube. This is a `list` of `two-tuples` where the first item is the dimension and the second item is the value to cut from. It is turned into a query where multible cuts for the same dimension are combined to an *OR* query and then the queries for the different dimensions are combined to an *AND* query. ``page`` Page the drilldown result and return page number *page*. type: `int` ``pagesize`` Page the drilldown result into page of size *pagesize*. type: `int` ``order`` Sort the result based on the dimension *sort_dimension*. This may be `None` (*default*) or a `list` of two-`tuples` where the first element is the *dimension* and the second element is the order (`False` for ascending, `True` for descending). Type: `list` of two-`tuples`. Raises: :exc:`ValueError` If a cube is not yet computed. Call :meth:`compute` to compute the cube. :exc:`KeyError` If a drilldown, cut or order dimension is not part of this cube or the order dimensions are not a subset of the drilldown dimensions. Returns: A `dict` containing the drilldown and the summary: {"drilldown": [ {"num_entries": 5545, "amount": 41087379002.0, "cofog1": {"description": "", "label": "Economic affairs"}}, ... ] "summary": {"amount": 7353306450299.0, "num_entries": 133612}} """ # Get the joins (aka alias) and the dataset joins = alias = self.alias dataset = self # Aggregation fields are all of the measures, so we create individual # summary fields with the sum function of SQLAlchemy fields = [db.func.sum(alias.c[m]).label(m) for m in measures] # We append an aggregation field that counts the number of entries fields.append(db.func.count(alias.c.id).label("entries")) # Create a copy of the statistics fields (for later) stats_fields = list(fields) # Create label map for time columns (year and month) for lookup # since they are found under the time attribute labels = { 'year': dataset['time']['year'].column_alias.label('year'), 'month': dataset['time']['yearmonth'].column_alias.label('month'), } # Get the dimensions we're interested in. These would be the drilldowns # and the cuts. For compound dimensions we are only interested in the # most significant one (e.g. for from.name we're interested in from) dimensions = drilldowns + [k for k, v in cuts] dimensions = [d.split('.')[0] for d in dimensions] # Loop over the dimensions as a set (to avoid multiple occurances) for dimension in set(dimensions): # If the dimension is year or month we're interested in 'time' if dimension in labels: dimension = 'time' # If the dimension table isn't in the automatic joins we add it if dimension not in [c.table.name for c in joins.columns]: joins = dataset[dimension].join(joins) # Drilldowns are performed using group_by SQL functions group_by = [] for key in drilldowns: # If drilldown is in labels we append its mapped column to fields if key in labels: column = labels[key] group_by.append(column) fields.append(column) else: # Get the column from the dataset column = dataset.key(key) # If the drilldown is a compound dimension or the columns table # is in the joins we're already fetching the column so we just # append it to fields and the group_by if '.' in key or column.table == alias: fields.append(column) group_by.append(column) else: # If not we add the column table to the fields and add all # of that tables columns to the group_by fields.append(column.table) for col in column.table.columns: group_by.append(col) # Cuts are managed using AND statements and we use a dict with set as # the default value to create the filters (cut on various values) conditions = db.and_() filters = defaultdict(set) for key, value in cuts: # If the key is in labels (year or month) we get the mapped column # else we get the column from the dataset if key in labels: column = labels[key] else: column = dataset.key(key) # We add the value to the set for that particular column filters[column].add(value) # Loop over the columns in the filter and add that to the conditions # For every value in the set we create and OR statement so we get e.g. # year=2007 AND (from.who == 'me' OR from.who == 'you') for attr, values in filters.items(): conditions.append(db.or_(*[attr == v for v in values])) # Ordering can be set by a parameter or ordered by measures by default order_by = [] # If no order is defined we default to order of the measures in the # order they occur (furthest to the left is most significant) if order is None or not len(order): order = [(m, True) for m in measures] # We loop through the order list to add the columns themselves for key, direction in order: # If it's a part of the measures we have to order by the # aggregated values (the sum of the measure) if key in measures: column = db.func.sum(alias.c[key]).label(key) # If it's in the labels we have to get the mapped column elif key in labels: column = labels[key] # ...if not we just get the column from the dataset else: column = dataset.key(key) # We append the column and set the direction (True == descending) order_by.append(column.desc() if direction else column.asc()) # query 1: get overall sums. # Here we use the stats_field we saved earlier query = db.select(stats_fields, conditions, joins) rp = dataset.bind.execute(query) # Execute the query and turn them to a list so we can pop the # entry count and then zip the measurements and the totals together stats = list(rp.fetchone()) num_entries = stats.pop() total = zip(measures, stats) # query 2: get total count of drilldowns if len(group_by): # Select 1 for each group in the group_by and count them query = db.select(['1'], conditions, joins, group_by=group_by) query = db.select([db.func.count('1')], '1=1', query.alias('q')) rp = dataset.bind.execute(query) num_drilldowns, = rp.fetchone() else: # If there are no drilldowns we still have to do one num_drilldowns = 1 # The drilldown result list drilldown = [] # The offset in the db, based on the page and pagesize (we have to # modify it since page counts starts from 1 but we count from 0 offset = int((page - 1) * pagesize) # query 3: get the actual data query = db.select(fields, conditions, joins, order_by=order_by, group_by=group_by, use_labels=True, limit=pagesize, offset=offset) rp = dataset.bind.execute(query) while True: # Get each row in the db result and append it, decoded, to the # drilldown result. The decoded version is a json represenation row = rp.fetchone() if row is None: break result = decode_row(row, dataset) drilldown.append(result) # Create the summary based on the stats_fields and other things # First we add a the total for each measurement in the root of the # summary (watch out!) and then we add various other, self-explanatory # statistics such as page, number of entries. The currency value is # strange since it's redundant for multiple measures but is left as is # for backwards compatibility summary = {key: value for (key, value) in total} summary.update({ 'num_entries': num_entries, 'currency': {m: dataset.currency for m in measures}, 'num_drilldowns': num_drilldowns, 'page': page, 'pages': int(math.ceil(num_drilldowns / float(pagesize))), 'pagesize': pagesize }) return {'drilldown': drilldown, 'summary': summary}
def aggregate(self, measures=['amount'], drilldowns=[], cuts=[], page=1, pagesize=10000, order=[]): """ Query the dataset for a subset of cells based on cuts and drilldowns. It returns a structure with a list of drilldown items and a summary about the slice cutted by the query. ``measures`` The numeric units to be aggregated over, defaults to [``amount``]. (type: `list`) ``drilldowns`` Dimensions to drill down to. (type: `list`) ``cuts`` Specification what to cut from the cube. This is a `list` of `two-tuples` where the first item is the dimension and the second item is the value to cut from. It is turned into a query where multible cuts for the same dimension are combined to an *OR* query and then the queries for the different dimensions are combined to an *AND* query. ``page`` Page the drilldown result and return page number *page*. type: `int` ``pagesize`` Page the drilldown result into page of size *pagesize*. type: `int` ``order`` Sort the result based on the dimension *sort_dimension*. This may be `None` (*default*) or a `list` of two-`tuples` where the first element is the *dimension* and the second element is the order (`False` for ascending, `True` for descending). Type: `list` of two-`tuples`. Raises: :exc:`ValueError` If a cube is not yet computed. Call :meth:`compute` to compute the cube. :exc:`KeyError` If a drilldown, cut or order dimension is not part of this cube or the order dimensions are not a subset of the drilldown dimensions. Returns: A `dict` containing the drilldown and the summary: {"drilldown": [ {"num_entries": 5545, "amount": 41087379002.0, "cofog1": {"description": "", "label": "Economic affairs"}}, ... ] "summary": {"amount": 7353306450299.0, "num_entries": 133612}} """ # Get the joins (aka alias) and the dataset joins = alias = self.alias dataset = self # Aggregation fields are all of the measures, so we create individual # summary fields with the sum function of SQLAlchemy fields = [func.sum(alias.c[m]).label(m) for m in measures] # We append an aggregation field that counts the number of entries fields.append(func.count(alias.c.id).label("entries")) # Create a copy of the statistics fields (for later) stats_fields = list(fields) # Create label map for time columns (year and month) for lookup # since they are found under the time attribute labels = { 'year': dataset['time']['year'].column_alias.label('year'), 'month': dataset['time']['yearmonth'].column_alias.label('month'), } # Get the dimensions we're interested in. These would be the drilldowns # and the cuts. For compound dimensions we are only interested in the # most significant one (e.g. for from.name we're interested in from) dimensions = drilldowns + [k for k, v in cuts] dimensions = [d.split('.')[0] for d in dimensions] # Loop over the dimensions as a set (to avoid multiple occurances) for dimension in set(dimensions): # If the dimension is year or month we're interested in 'time' if dimension in labels: dimension = 'time' # If the dimension table isn't in the automatic joins we add it if dimension not in [c.table.name for c in joins.columns]: joins = dataset[dimension].join(joins) # Drilldowns are performed using group_by SQL functions group_by = [] for key in drilldowns: # If drilldown is in labels we append its mapped column to fields if key in labels: column = labels[key] group_by.append(column) fields.append(column) else: # Get the column from the dataset column = dataset.key(key) # If the drilldown is a compound dimension or the columns table # is in the joins we're already fetching the column so we just # append it to fields and the group_by if '.' in key or column.table == alias: fields.append(column) group_by.append(column) else: # If not we add the column table to the fields and add all # of that tables columns to the group_by fields.append(column.table) for col in column.table.columns: group_by.append(col) # Cuts are managed using AND statements and we use a dict with set as # the default value to create the filters (cut on various values) conditions = and_() filters = defaultdict(set) for key, value in cuts: # If the key is in labels (year or month) we get the mapped column # else we get the column from the dataset if key in labels: column = labels[key] else: column = dataset.key(key) # We add the value to the set for that particular column filters[column].add(value) # Loop over the columns in the filter and add that to the conditions # For every value in the set we create and OR statement so we get e.g. # year=2007 AND (from.who == 'me' OR from.who == 'you') for attr, values in filters.items(): conditions.append(or_(*[attr == v for v in values])) # Ordering can be set by a parameter or ordered by measures by default order_by = [] # If no order is defined we default to order of the measures in the # order they occur (furthest to the left is most significant) if order is None or not len(order): order = [(m, True) for m in measures] # We loop through the order list to add the columns themselves for key, direction in order: # If it's a part of the measures we have to order by the # aggregated values (the sum of the measure) if key in measures: column = func.sum(alias.c[key]).label(key) # If it's in the labels we have to get the mapped column elif key in labels: column = labels[key] # ...if not we just get the column from the dataset else: column = dataset.key(key) # We append the column and set the direction (True == descending) order_by.append(column.desc() if direction else column.asc()) # query 1: get overall sums. # Here we use the stats_field we saved earlier query = select(stats_fields, conditions, joins) rp = dataset.bind.execute(query) # Execute the query and turn them to a list so we can pop the # entry count and then zip the measurements and the totals together stats = list(rp.fetchone()) num_entries = stats.pop() total = zip(measures, stats) # query 2: get total count of drilldowns if len(group_by): # Select 1 for each group in the group_by and count them query = select(['1'], conditions, joins, group_by=group_by) query = select([func.count('1')], '1=1', query.alias('q')) rp = dataset.bind.execute(query) num_drilldowns, = rp.fetchone() else: # If there are no drilldowns we still have to do one num_drilldowns = 1 # The drilldown result list drilldown = [] # The offset in the db, based on the page and pagesize (we have to # modify it since page counts starts from 1 but we count from 0 offset = int((page - 1) * pagesize) # query 3: get the actual data query = select(fields, conditions, joins, order_by=order_by, group_by=group_by, use_labels=True, limit=pagesize, offset=offset) rp = dataset.bind.execute(query) while True: # Get each row in the db result and append it, decoded, to the # drilldown result. The decoded version is a json represenation row = rp.fetchone() if row is None: break result = decode_row(row, dataset) drilldown.append(result) # Create the summary based on the stats_fields and other things # First we add a the total for each measurement in the root of the # summary (watch out!) and then we add various other, self-explanatory # statistics such as page, number of entries. The currency value is # strange since it's redundant for multiple measures but is left as is # for backwards compatibility summary = {key: value for (key, value) in total} summary.update({ 'num_entries': num_entries, 'currency': {m: dataset.currency for m in measures}, 'num_drilldowns': num_drilldowns, 'page': page, 'pages': int(math.ceil(num_drilldowns / float(pagesize))), 'pagesize': pagesize }) return {'drilldown': drilldown, 'summary': summary}
def aggregate(self, measure='amount', drilldowns=None, cuts=None, page=1, pagesize=10000, order=None): """ Query the dataset for a subset of cells based on cuts and drilldowns. It returns a structure with a list of drilldown items and a summary about the slice cutted by the query. ``measure`` The numeric unit to be aggregated over, defaults to ``amount``. ``drilldowns`` Dimensions to drill down to. (type: `list`) ``cuts`` Specification what to cut from the cube. This is a `list` of `two-tuples` where the first item is the dimension and the second item is the value to cut from. It is turned into a query where multible cuts for the same dimension are combined to an *OR* query and then the queries for the different dimensions are combined to an *AND* query. ``page`` Page the drilldown result and return page number *page*. type: `int` ``pagesize`` Page the drilldown result into page of size *pagesize*. type: `int` ``order`` Sort the result based on the dimension *sort_dimension*. This may be `None` (*default*) or a `list` of two-`tuples` where the first element is the *dimension* and the second element is the order (`False` for ascending, `True` for descending). Type: `list` of two-`tuples`. Raises: :exc:`ValueError` If a cube is not yet computed. Call :meth:`compute` to compute the cube. :exc:`KeyError` If a drilldown, cut or order dimension is not part of this cube or the order dimensions are not a subset of the drilldown dimensions. Returns: A `dict` containing the drilldown and the summary:: {"drilldown": [ {"num_entries": 5545, "amount": 41087379002.0, "cofog1": {"description": "", "label": "Economic affairs"}}, ... ] "summary": {"amount": 7353306450299.0, "num_entries": 133612}} """ cuts = cuts or [] drilldowns = drilldowns or [] order = order or [] joins = alias = self.alias dataset = self fields = [db.func.sum(alias.c[measure]).label(measure), db.func.count(alias.c.id).label("entries")] stats_fields = list(fields) labels = { 'year': dataset['time']['year'].column_alias.label('year'), 'month': dataset['time']['yearmonth'].column_alias.label('month'), } dimensions = drilldowns + [k for k, v in cuts] dimensions = [d.split('.')[0] for d in dimensions] for dimension in set(dimensions): if dimension in labels: dimension = 'time' if dimension not in [c.table.name for c in joins.columns]: joins = dataset[dimension].join(joins) group_by = [] for key in drilldowns: if key in labels: column = labels[key] group_by.append(column) fields.append(column) else: column = dataset.key(key) if '.' in key or column.table == alias: fields.append(column) group_by.append(column) else: fields.append(column.table) for col in column.table.columns: group_by.append(col) conditions = db.and_() filters = defaultdict(set) for key, value in cuts: if key in labels: column = labels[key] else: column = dataset.key(key) filters[column].add(value) for attr, values in filters.items(): conditions.append(db.or_(*[attr == v for v in values])) order_by = [] if order is None or not len(order): order = [(measure, True)] for key, direction in order: if key == measure: column = db.func.sum(alias.c[measure]).label(measure) elif key in labels: column = labels[key] else: column = dataset.key(key) order_by.append(column.desc() if direction else column.asc()) # query 1: get overall sums. query = db.select(stats_fields, conditions, joins) rp = dataset.bind.execute(query) total, num_entries = rp.fetchone() # query 2: get total count of drilldowns if len(group_by): query = db.select(['1'], conditions, joins, group_by=group_by) query = db.select([db.func.count('1')], '1=1', query.alias('q')) rp = dataset.bind.execute(query) num_drilldowns, = rp.fetchone() else: num_drilldowns = 1 drilldown = [] offset = int((page - 1) * pagesize) # query 3: get the actual data query = db.select(fields, conditions, joins, order_by=order_by, group_by=group_by, use_labels=True, limit=pagesize, offset=offset) rp = dataset.bind.execute(query) while True: row = rp.fetchone() if row is None: break result = decode_row(row, dataset) drilldown.append(result) return { 'drilldown': drilldown, 'summary': { measure: total, 'num_entries': num_entries, 'currency': {measure: dataset.currency}, 'num_drilldowns': num_drilldowns, 'page': page, 'pages': int(math.ceil(num_drilldowns / float(pagesize))), 'pagesize': pagesize } }
def aggregate(self, measure='amount', drilldowns=None, cuts=None, page=1, pagesize=10000, order=None): """ Query the dataset for a subset of cells based on cuts and drilldowns. It returns a structure with a list of drilldown items and a summary about the slice cutted by the query. ``measure`` The numeric unit to be aggregated over, defaults to ``amount``. ``drilldowns`` Dimensions to drill down to. (type: `list`) ``cuts`` Specification what to cut from the cube. This is a `list` of `two-tuples` where the first item is the dimension and the second item is the value to cut from. It is turned into a query where multible cuts for the same dimension are combined to an *OR* query and then the queries for the different dimensions are combined to an *AND* query. ``page`` Page the drilldown result and return page number *page*. type: `int` ``pagesize`` Page the drilldown result into page of size *pagesize*. type: `int` ``order`` Sort the result based on the dimension *sort_dimension*. This may be `None` (*default*) or a `list` of two-`tuples` where the first element is the *dimension* and the second element is the order (`False` for ascending, `True` for descending). Type: `list` of two-`tuples`. Raises: :exc:`ValueError` If a cube is not yet computed. Call :meth:`compute` to compute the cube. :exc:`KeyError` If a drilldown, cut or order dimension is not part of this cube or the order dimensions are not a subset of the drilldown dimensions. Returns: A `dict` containing the drilldown and the summary:: {"drilldown": [ {"num_entries": 5545, "amount": 41087379002.0, "cofog1": {"description": "", "label": "Economic affairs"}}, ... ] "summary": {"amount": 7353306450299.0, "num_entries": 133612}} """ cuts = cuts or [] drilldowns = drilldowns or [] order = order or [] joins = alias = self.alias dataset = self fields = [ db.func.sum(alias.c[measure]).label(measure), db.func.count(alias.c.id).label("entries") ] stats_fields = list(fields) labels = { 'year': dataset['time']['year'].column_alias.label('year'), 'month': dataset['time']['yearmonth'].column_alias.label('month'), } dimensions = drilldowns + [k for k, v in cuts] dimensions = [d.split('.')[0] for d in dimensions] for dimension in set(dimensions): if dimension in labels: dimension = 'time' if dimension not in [c.table.name for c in joins.columns]: joins = dataset[dimension].join(joins) group_by = [] for key in drilldowns: if key in labels: column = labels[key] group_by.append(column) fields.append(column) else: column = dataset.key(key) if '.' in key or column.table == alias: fields.append(column) group_by.append(column) else: fields.append(column.table) for col in column.table.columns: group_by.append(col) conditions = db.and_() filters = defaultdict(set) for key, value in cuts: if key in labels: column = labels[key] else: column = dataset.key(key) filters[column].add(value) for attr, values in filters.items(): conditions.append(db.or_(*[attr == v for v in values])) order_by = [] if order is None or not len(order): order = [(measure, True)] for key, direction in order: if key == measure: column = db.func.sum(alias.c[measure]).label(measure) elif key in labels: column = labels[key] else: column = dataset.key(key) order_by.append(column.desc() if direction else column.asc()) # query 1: get overall sums. query = db.select(stats_fields, conditions, joins) rp = dataset.bind.execute(query) total, num_entries = rp.fetchone() # query 2: get total count of drilldowns if len(group_by): query = db.select(['1'], conditions, joins, group_by=group_by) query = db.select([db.func.count('1')], '1=1', query.alias('q')) rp = dataset.bind.execute(query) num_drilldowns, = rp.fetchone() else: num_drilldowns = 1 drilldown = [] offset = int((page - 1) * pagesize) # query 3: get the actual data query = db.select(fields, conditions, joins, order_by=order_by, group_by=group_by, use_labels=True, limit=pagesize, offset=offset) rp = dataset.bind.execute(query) while True: row = rp.fetchone() if row is None: break result = decode_row(row, dataset) drilldown.append(result) return { 'drilldown': drilldown, 'summary': { measure: total, 'num_entries': num_entries, 'currency': { measure: dataset.currency }, 'num_drilldowns': num_drilldowns, 'page': page, 'pages': int(math.ceil(num_drilldowns / float(pagesize))), 'pagesize': pagesize } }