def dp_query_test(self,
d1_query,
d2_query,
debug=False,
plot=True,
bound=True,
exact=False,
repeat_count=10000,
confidence=0.95,
get_exact=True):
ag = agg.Aggregation(t=1, repeat_count=repeat_count)
d1, d2, d1_metadata, d2_metadata = self.generate_neighbors(
load_csv=True)
fD1, fD1_actual, fD1_low, fD1_high = ag.run_agg_query(
d1, d1_metadata, d1_query, confidence, get_exact)
fD2, fD2_actual, fD2_low, fD2_high = ag.run_agg_query(
d2, d2_metadata, d2_query, confidence, get_exact)
d1hist, d2hist, bin_edges = self.generate_histogram_neighbors(
fD1, fD2, binsize="auto")
d1size, d2size = fD1.size, fD2.size
dp_res, d1histupperbound, d2histupperbound, d1lower, d2lower = self.dp_test(
d1hist, d2hist, bin_edges, d1size, d2size, debug)
acc_res, utility_res, within_bounds = self.accuracy_test(
fD1_actual, fD1_low, fD1_high, confidence)
bias_res, msd = self.bias_test(fD1_actual, fD1)
if (plot):
self.plot_histogram_neighbors(fD1, fD2, d1histupperbound,
d2histupperbound, d1hist, d2hist,
d1lower, d2lower, bin_edges, bound,
exact)
return dp_res, acc_res, utility_res, bias_res
def dp_powerset_test(self,
query_str,
debug=False,
plot=True,
bound=True,
exact=False,
repeat_count=10000,
confidence=0.95,
test_cases=5):
ag = agg.Aggregation(t=1, repeat_count=repeat_count)
ex = exp.Exploration()
res_list = {}
halton_samples = ex.generate_halton_samples(bounds=ex.corners,
dims=ex.N,
n_sample=test_cases)
# Iterate through each sample generated by halton sequence
for sample in halton_samples:
df, metadata = ex.create_small_dataset(sample)
ex.generate_powerset(df)
print("Test case: ", list(sample))
for filename in ex.visited:
print("Testing: ", filename)
d1_query = query_str + "d1_" + filename + "." + "d1_" + filename
d2_query = query_str + "d2_" + filename + "." + "d2_" + filename
[d1, d2, d1_metadata, d2_metadata] = ex.neighbor_pair[filename]
fD1, fD1_actual, fD1_low, fD1_high = ag.run_agg_query(
d1, d1_metadata, d1_query, confidence)
fD2, fD2_actual, fD2_low, fD2_high = ag.run_agg_query(
d2, d2_metadata, d2_query, confidence)
#acc_res, utility_res, within_bounds = self.accuracy_test(fD1_actual, fD1_low, fD1_high, confidence)
acc_res, utility_res, within_bounds = None, None, None
bias_res, msd = self.bias_test(fD1_actual, fD1)
d1hist, d2hist, bin_edges = self.generate_histogram_neighbors(
fD1, fD2, binsize="auto")
d1size, d2size = fD1.size, fD2.size
dp_res, d1histupperbound, d2histupperbound, d1lower, d2lower = self.dp_test(
d1hist, d2hist, bin_edges, d1size, d2size, debug)
print("DP Predicate Test Result: ", dp_res)
if (plot):
self.plot_histogram_neighbors(fD1, fD2, d1histupperbound,
d2histupperbound, d1hist,
d2hist, d1lower, d2lower,
bin_edges, bound, exact)
key = "[" + ','.join(str(e)
for e in list(sample)) + "] - " + filename
res_list[key] = [
dp_res, acc_res, utility_res, within_bounds, bias_res, msd
]
print("Halton sequence based Powerset Test Result")
for data, res in res_list.items():
print(data, "-", res)
dp_res = np.all(np.array([res[0] for data, res in res_list.items()]))
#acc_res = np.all(np.array([res[1] for res in res_list]))
#utility_res = np.all(np.array([res[2] for res in res_list]))
acc_res, utility_res = None, None
bias_res = np.all(np.array([res[4] for data, res in res_list.items()]))
return dp_res, acc_res, utility_res, bias_res
def whitenoise_core_test(self,
dataset_path,
col_names,
f,
*args,
numbins=0,
binsize="auto",
debug=False,
plot=True,
bound=True,
exact=False,
repeat_count=100,
epsilon=1.0,
actual=1.0,
**kwargs):
ag = agg.Aggregation(t=1, repeat_count=repeat_count)
self.dataset_path = dataset_path
d1, d2, d1_metadata, d2_metadata = self.generate_neighbors(
load_csv=True)
d1_file_path = os.path.join(self.file_dir, self.csv_path, "d1.csv")
d2_file_path = os.path.join(self.file_dir, self.csv_path, "d2.csv")
if (len(args) == 3):
fD1 = ag.whitenoise_core_dp_multi_agg(f, d1_file_path, col_names,
args, epsilon, kwargs)
fD2 = ag.whitenoise_core_dp_multi_agg(f, d2_file_path, col_names,
args, epsilon, kwargs)
else:
fD1 = ag.whitenoise_core_dp_agg(f, d1_file_path, col_names, args,
epsilon, kwargs)
fD2 = ag.whitenoise_core_dp_agg(f, d2_file_path, col_names, args,
epsilon, kwargs)
d1size, d2size = fD1.size, fD2.size
d1hist, d2hist, bin_edges = \
self.generate_histogram_neighbors(fD1, fD2, numbins, binsize, exact=exact)
dp_res, d1histupperbound, d2histupperbound, d1lower, d2lower = self.dp_test(
d1hist, d2hist, bin_edges, d1size, d2size, debug)
print("DP Predicate Test:", dp_res, "\n")
bias_res, msd = self.bias_test(actual, fD1)
print("Bias Test:", bias_res, "\n")
if (plot):
self.plot_histogram_neighbors(fD1, fD2, d1histupperbound,
d2histupperbound, d1hist, d2hist,
d1lower, d2lower, bin_edges, bound)
return dp_res, bias_res
def dp_groupby_query_test(self,
d1_query,
d2_query,
debug=False,
plot=True,
bound=True,
exact=False,
repeat_count=10000,
confidence=0.95):
ag = agg.Aggregation(t=1, repeat_count=repeat_count)
d1, d2, d1_metadata, d2_metadata = self.generate_neighbors(
load_csv=True)
d1_res, d1_exact, dim_cols, num_cols = ag.run_agg_query_df(
d1, d1_metadata, d1_query, confidence, file_name="d1")
d2_res, d2_exact, dim_cols, num_cols = ag.run_agg_query_df(
d2, d2_metadata, d2_query, confidence, file_name="d2")
res_list = []
for col in num_cols:
d1_gp = d1_res.groupby(dim_cols)[col].apply(list).reset_index(
name=col)
d2_gp = d2_res.groupby(dim_cols)[col].apply(list).reset_index(
name=col)
exact_gp = d1_exact.groupby(dim_cols)[col].apply(list).reset_index(
name=col)
# Full outer join after flattening the results above to one row per dimension key
# We cannot be sure if every dimension key has a response in every repeated query run because of tau thresholding
# That's why we do a full outer join and flatten whatever vector of results we get for the numerical column across repeat runs
# This is what we use for generating the histogram of results for that dimension key
d1_d2 = d1_gp.merge(d2_gp, on=dim_cols, how='outer')
d1_d2 = d1_d2.merge(exact_gp, on=dim_cols, how='left')
n_cols = len(d1_d2.columns)
for index, row in d1_d2.iterrows():
print(d1_d2.iloc[index, :n_cols - 3])
print("Column: ", col)
# fD1 and fD2 will have the results of the K repeated query results that can be passed through histogram test
# These results are for that particular numerical column and the specific dimension key of d1_d2
fD1 = np.array(
[val[0] for val in d1_d2.iloc[index, n_cols - 3]])
fD2 = np.array(
[val[0] for val in d1_d2.iloc[index, n_cols - 2]])
exact_val = d1_d2.iloc[index, n_cols - 1][0]
d1hist, d2hist, bin_edges = self.generate_histogram_neighbors(
fD1, fD2, binsize="auto")
d1size, d2size = fD1.size, fD2.size
dp_res, d1histupperbound, d2histupperbound, d1lower, d2lower = self.dp_test(
d1hist, d2hist, bin_edges, d1size, d2size, debug)
print("DP Predicate Test Result: ", dp_res)
# Accuracy Test
low = np.array(
[val[1] for val in d1_d2.iloc[index, n_cols - 2]])
high = np.array(
[val[2] for val in d1_d2.iloc[index, n_cols - 2]])
acc_res, utility_res, within_bounds = self.accuracy_test(
exact_val, low, high, confidence)
bias_res, msd = self.bias_test(exact_val, fD1)
res_list.append([
dp_res, acc_res, utility_res, within_bounds, bias_res, msd
])
if (plot):
self.plot_histogram_neighbors(fD1, fD2, d1histupperbound,
d2histupperbound, d1hist,
d2hist, d1lower, d2lower,
bin_edges, bound, exact)
for res in res_list:
print(res)
res_list = res_list.values() if hasattr(
res_list, "values") else res_list # TODO why is this needed?
dp_res = np.all(np.array([res[0] for res in res_list]))
acc_res = np.all(np.array([res[1] for res in res_list]))
utility_res = np.all(np.array([res[2] for res in res_list]))
bias_res = np.all(np.array([res[4] for res in res_list]))
return dp_res, acc_res, utility_res, bias_res