def test_3():
"""
The result table without the k-condition is hardcoded for
the query -
Top-4 Salespersons with maximum Sales
Here the top-k vs others suggestion is asserted when the
ratio = topk_sum / others_sum is negative
"""
# hardcoding the artificial results table
result_table = pandas.DataFrame()
result_table['Salesperson'] = pandas.Series(['A', 'B', 'C', 'D', 'E', 'F', 'G'])
result_table['Sales'] = pandas.Series([5, 4, 3, 3, -1, -2, -10])
k = 4
metric = 'Sales'
# The data should be already sorted
assert(result_table[metric].is_monotonic_decreasing)
# suggestions = oversights.looking_at_tails.looking_at_tails(result_table, k, metric)
suggestions = topk_vs_others.topk_vs_others(result_table, k, metric)
print(suggestions)
expected_suggestions = """{'change_list': {'topKLimit': 7}, 'suggestion': 'The sum of Sales in top-k rows is negative whereas sum of rows not in top-k is positive', 'confidence_score': -1.1538461538461537}"""
assert(expected_suggestions == str(suggestions))
def test_2():
"""
The result table without the k-condition is hardcoded for
the query -
Top-4 Salespersons with maximum Sales
Here the top-k vs others suggestion is asserted when the
ratio = topk_sum / others_sum is positive and is more than
the threshold, so no suggestion is returned
"""
# hardcoding the artificial results table
result_table = pandas.DataFrame()
result_table['Salesperson'] = pandas.Series(['A', 'B', 'C', 'D', 'E', 'F', 'G'])
result_table['Sales'] = pandas.Series([5, 4, 3, 3, 3, 2, 1])
k = 4
metric = 'Sales'
# The data should be already sorted
assert(result_table[metric].is_monotonic_decreasing)
# suggestions = oversights.looking_at_tails.looking_at_tails(result_table, k, metric)
suggestions = topk_vs_others.topk_vs_others(result_table, k, metric)
print(suggestions)
expected_suggestions = """None"""
assert(expected_suggestions == str(suggestions))
def test_1():
"""
Tests the topk_vs_others based on the result table generated & stored
in data/data_for_test_topk_vs_others/result_table for the query -
Top-10 Item based on minimum unit costs.
Here the top-k vs others suggestion is asserted when the
ratio = topk_sum / others_sum is positive and is less than
the threshold
"""
result_table = pandas.read_csv(
'data/data_for_test_topk_vs_others/result_table.csv')
k = 10
metric = 'UnitCost'
# The data should be already sorted
assert (result_table[metric].is_monotonic_increasing)
# suggestions = oversights.looking_at_tails.looking_at_tails(result_table, k, metric)
suggestions = topk_vs_others.topk_vs_others(result_table, k, metric)
print(suggestions)
expected_suggestions = """{'oversight': <Oversights.TOPK_VS_OTHERS: 6>, 'change_list': {'topKLimit': 43}, 'suggestion': 'The rows NOT in the top-k have a much larger sum over UnitCost than the rows in top-k', 'confidence_score': 0.008621645877239863}"""
assert (expected_suggestions == str(suggestions))
def topk(table, metric, dimensions, is_asc, k, **kwargs):
""" This function returns both the results according to the intent
as well as the debiasing suggestions.
Also, if summary operator is applied, the name of metric column is
renamed to "<summary operator> of metric".
Oversights that may be detected in top-k
1. Regression to the mean
2. Looking at tails to find causes
3. Duplicates in top-k
4. More than just top-k
5. Top-k vs others
6. Top-k when less than k present
Args:
table: Type-pandas.dataframe
It has the contents of the csv file
metric: Type-string
It is the name of the column according to which we sort,
and in the case when grouping has to be done,
summary operator is applied on metric. Metric could a column
containing strings, if we are applying count operator on it.
dimensions: Type-list of str
It is the name of column we want.
In query:'top 5 batsman according to runs', dimension is 'batsman'.
When summary_operator is not None, we group by dimensions.
is_asc: Type-Bool
Denotes the sort order, True for ascending, False for Descending
k: Type-int
It is the number of entries to be taken
date_range: Type-tuple
Tuple of start_date and end_date
date_column_name: Type-str
It is the name of column which contains date
day_first: Type-str
Day_first denotes that does day in the date occurs before month in the
dates in the date column
Example - '29-02-19', here day_first is true
slices: Type-List of tuples
Tuple represents the conditon to keep the row.
(column_name, filter, value)
column_name - is the value of the column that the
condition is applied upon.
filter - Filters enum members, ex. Filters.IN
summary_operator: Type-summary_operators enum members
It denotes the summary operator, after grouping by dimensions.
ex. SummaryOperators.MAX, SummaryOperators.SUM
Note-summary_operator is always applied on metric column passed,
and only when grouping is done
Returns:
The function will return both suggestions and the results in a tuple.
(results, suggestions)
results: Type -pandas dataframe, The results of the weighted mean intent
suggestions: Type - List of dictionaries(suggestion structure), List of
suggestions.
"""
date_column_name = kwargs.get('date_column_name', 'date')
date_range = kwargs.get('date_range', None)
day_first = kwargs.get('day_first', False)
slices = kwargs.get('slices', None)
summary_operator = kwargs.get('summary_operator', None)
result_tuple = topk_results(table,
metric,
dimensions,
is_asc,
k,
date_column_name=date_column_name,
date_range=date_range,
day_first=day_first,
slices=slices,
summary_operator=summary_operator)
result_table = result_tuple[0]
suggestions = result_tuple[1]
duplicates_in_topk_suggestion = duplicates_in_topk(result_table,
dimensions)
if duplicates_in_topk_suggestion is not None:
suggestions.append(duplicates_in_topk_suggestion)
else:
# Check for RMT suggestion only when no duplicates present.
rmt_suggestion = regression_to_mean(table,
metric,
dimensions,
is_asc,
k,
date_column_name=date_column_name,
date_range=date_range,
day_first=day_first,
slices=slices,
summary_operator=summary_operator)
if rmt_suggestion is not None:
suggestions.append(rmt_suggestion)
results_without_k_condition = topk_results(
table,
metric,
dimensions,
is_asc,
-1,
date_column_name=date_column_name,
date_range=date_range,
day_first=day_first,
slices=slices,
summary_operator=summary_operator)[0]
more_than_just_topk_suggestion = more_than_just_topk(
results_without_k_condition, k, metric)
if more_than_just_topk_suggestion is not None:
suggestions.append(more_than_just_topk_suggestion)
topk_vs_others_suggestion = topk_vs_others(results_without_k_condition, k,
metric)
if topk_vs_others_suggestion is not None:
suggestions.append(topk_vs_others_suggestion)
looking_at_tails_suggestion = looking_at_tails(results_without_k_condition,
k, metric)
if looking_at_tails_suggestion is not None:
suggestions.append(looking_at_tails_suggestion)
topk_when_less_than_k_present_suggestion = topk_when_less_than_k_present(
result_table, k)
if topk_when_less_than_k_present_suggestion is not None:
suggestions.append(topk_when_less_than_k_present_suggestion)
order = oversights_order.ORDER_IN_TOPK
suggestions = rank_oversights.rank_oversights(suggestions, order)
if summary_operator is not None:
result_table = aspects.update_metric_column_name(
result_table, summary_operator, metric)
return (result_table, suggestions)
def topk(table, metric, dimensions, is_asc, k, **kwargs):
""" This function returns both the results according to the intent
as well as the debiasing suggestions.
Some of the oversights considered in this intent are-
1. Regression to the mean
2. Looking at tails to find causes - TODO
Args:
table: Type-pandas.dataframe
It has the contents of the csv file
metric: Type-string
It is the name of the column according to which we sort,
and in the case when grouping has to be done,
summary operator is applied on metric. Metric could a column
containing strings, if we are applying count operator on it.
dimensions: Type-list of str
It is the name of column we want.
In query:'top 5 batsman according to runs', dimension is 'batsman'.
When summary_operator is not None, we group by dimensions.
is_asc: Type-Bool
Denotes the sort order, True for ascending, False for Descending
k: Type-int
It is the number of entries to be taken
date_range: Type-tuple
Tuple of start_date and end_date
date_column_name: Type-str
It is the name of column which contains date
date_format: Type-str
It is required by datetime.strp_time to parse the date in the format
Format Codes
https://docs.python.org/3/library/datetime.html#strftime-and-strptime-behavior
slices: Type-List of tuples
Tuple represents the conditon to keep the row.
(column_name, filter, value)
column_name - is the value of the column that the
condition is applied upon.
filter - Filters enum members, ex. Filters.IN
summary_operator: Type-summary_operators enum members
It denotes the summary operator, after grouping by dimensions.
ex. SummaryOperators.MAX, SummaryOperators.SUM
Note-summary_operator is always applied on metric column passed,
and only when grouping is done
Returns:
The function will return both suggestions and the results in a tuple.
(results, suggestions)
results: Type - pandas dataframe, The results of the intended top-k
suggestions: Type - List of strings, List of suggestions.
"""
date_column_name = kwargs.get('date_column_name', 'date')
date_range = kwargs.get('date_range', None)
date_format = kwargs.get('date_format', '%Y-%m-%d')
slices = kwargs.get('slices', None)
summary_operator = kwargs.get('summary_operator', None)
result_table = topk_results(table,
metric,
dimensions,
is_asc,
k,
date_column_name=date_column_name,
date_range=date_range,
date_format=date_format,
slices=slices,
summary_operator=summary_operator)
suggestions = []
duplicates_in_topk_suggestion = duplicates_in_topk(result_table,
dimensions)
if duplicates_in_topk_suggestion is not None:
suggestions.append(duplicates_in_topk_suggestion)
else:
# Check for RMT suggestion only when no duplicates present.
rmt_suggestion = regression_to_mean(table,
metric,
dimensions,
is_asc,
k,
date_column_name=date_column_name,
date_range=date_range,
date_format=date_format,
slices=slices,
summary_operator=summary_operator)
if rmt_suggestion is not None:
suggestions.append(rmt_suggestion)
results_without_k_condition = topk_results(
table,
metric,
dimensions,
is_asc,
-1,
date_column_name=date_column_name,
date_range=date_range,
date_format=date_format,
slices=slices,
summary_operator=summary_operator)
more_than_just_topk_suggestion = more_than_just_topk(
results_without_k_condition, k, metric)
if more_than_just_topk_suggestion is not None:
suggestions.append(more_than_just_topk_suggestion)
topk_vs_others_suggestion = topk_vs_others(results_without_k_condition, k,
metric)
if topk_vs_others_suggestion is not None:
suggestions.append(topk_vs_others_suggestion)
looking_at_tails_suggestion = looking_at_tails(results_without_k_condition,
k, metric)
if looking_at_tails_suggestion is not None:
suggestions.append(looking_at_tails_suggestion)
return (result_table, suggestions)