def get_rel_responder_abs_df(treatment, con, dfs=None, use_percentage=None, use_labels=None):
if SELECTION != "resp":
return
df_prop, df_resp = get_prop_resp(treatment)
df_prop[df_resp.columns] = df_resp
df_prop_full, df_resp_ref = get_prop_resp("t12a")
resp_values = metrics.get_data(metrics.get_rel_responder_abs_df(df_prop))
resp_ref_values = metrics.get_data(metrics.get_rel_responder_abs_df(df_prop_full))
table, res = rp.ttest(pd.Series(resp_values), pd.Series(resp_ref_values), paired=False)
s = res.results[2]
p = res.results[3]
r = res.results[9]
diff = res.results[0]
dof = res.results[1]
s = res.results[2]
p = res.results[3]
r = res.results[9]
print("Conclusion: ", generate_stat_sentence(np.mean(resp_ref_values), np.std(resp_ref_values), np.mean(resp_values), np.std(resp_values), s, p, dof, diff=diff, label1="t12.dss", label2=treatment+".dss"))
print("Table:", table)
print("Res:", res)
res = {
"rel. min_offer T12": metrics.get_mean(resp_ref_values),
"rel. min_offer T13": metrics.get_mean(resp_values),
# "rejection_ratio": rejection_ratio(df_prop)
}
test_label = f"(ttest independent) H0: equal"
res = {k: (f"{v:.3f}" if pd.notnull(v) and v!= int(v) else v) for k,v in res.items()}
res["min_offer" + test_label] = f"{s:.3f} (p: {p:.3f}, r: {r:.3f})"
print()
return res
import os
import os.path
import numpy as np
import xgboost as xgb
from metrics import F1, get_data, apply_PCA, feature_selection
if __name__ == "__main__":
training_samples, validation_samples, test_samples, training_labels, validation_labels = get_data(
)
# training_samples = feature_selection(training_samples, training_labels)
# validation_samples = feature_selection(validation_samples, validation_labels)
# test_samples = feature_selection(test_samples)
# training_samples = apply_PCA(training_samples)
# validation_samples = apply_PCA(validation_samples)
# test_samples = apply_PCA(test_samples)
dtrain = xgb.DMatrix(training_samples, training_labels)
# dtest = xgb.DMatrix(validation_samples)
dtest = xgb.DMatrix(test_samples)
# max_depths = [32, 64, 128, 256, 512]
# num_rounds = [32, 64, 128, 256, 512]
# learning_rates = [0.8, 1]
max_depths = [64]
num_rounds = [64]
learning_rates = [0.8]
for max_depth in max_depths:
for num_round in num_rounds:
for learning_rate in learning_rates:
param = {
'max_depth': max_depth,
'eta': learning_rate,
'silent': 1,
def _get_prop_vs_prop_dss_score(treatment,
con,
dfs=None,
use_percentage=None,
use_labels=None,
metric=None,
as_percentage=None,
is_categorical=None,
alternative=None):
df_prop, df_resp = get_prop_resp(treatment)
df_prop_ref, df_resp_ref = get_prop_resp("t11a")
print(metric.__name__)
metric_values = metric(df_prop)
metric_value = metrics.get_mean(metric_values)
metric_ref_values = metric(df_prop_ref)
metric_value_ref = metrics.get_mean(metric_ref_values)
metric_values = metrics.get_data(metric_values)
metric_ref_values = metrics.get_data(metric_ref_values)
dof = 0
diff = None
if is_categorical:
table, res = rp.crosstab(pd.Series(metric_ref_values),
pd.Series(metric_values),
test='chi-square')
s, p, r = res.results.values
print(
"Conclusion: ",
generate_cat_stat_sentence(np.mean(metric_ref_values),
np.std(metric_ref_values),
np.mean(metric_values),
np.std(metric_values),
s,
p,
dof,
diff=diff,
label1="t11a.dss",
label2=treatment + ".dss"))
test_label = f"(pearson chi2)"
else:
#print("Ranksums", stats.ranksums(metric_ref_values, metric_values))
table, res = rp.ttest(pd.Series(metric_ref_values),
pd.Series(metric_values),
paired=False)
s = res.results[2]
if alternative == "greater":
p = res.results[4]
elif alternative == "less":
p = res.results[5]
elif alternative in (None, 'two-sided'):
p = res.results[3]
r = res.results[9]
diff = res.results[0]
dof = res.results[1]
s = res.results[2]
p = res.results[3]
r = res.results[9]
print(
"Conclusion: ",
generate_stat_sentence(np.mean(metric_ref_values),
np.std(metric_ref_values),
np.mean(metric_values),
np.std(metric_values),
s,
p,
dof,
diff=diff,
label1="t11a.dss",
label2=treatment + ".dss"))
test_label = f"(ttest independent) H0: {'equal' if alternative in {None, 'two-sided'} else alternative}"
print("RESUME: ", res)
print("TABLE: ", table)
if as_percentage:
res = {
# "Proposer": f'{100 * prop_value:.2f} %',
"Proposer + DSS":
f'{100 * metric_value:.2f} %',
"T11A ":
f'{100 * metric_value_ref:.2f} %',
"prop:dss - auto prop":
f'{100 * (metric_value - metric_value_ref):.2f} %',
}
else:
res = {
# "Proposer": f'{prop_value:.2f}',
"Proposer + DSS": f'{metric_value:.2f}',
"T11A": f'{metric_value_ref:.2f}',
"prop:dss - auto prop":
f'{(metric_value - metric_value_ref):.2f} %',
}
res[test_label] = f"{s:.3f} ({p:.3f})"
if is_categorical:
res[test_label] = f"{s:.3f} (p: {p:.3f}, phi: {r:.3f})"
else:
res[test_label] = f"{s:.3f} (p: {p:.3f}, r: {r:.3f})"
return res
def _get_prop_vs_prop_dss_score(treatment,
con,
dfs=None,
use_percentage=None,
use_labels=None,
metric=None,
as_percentage=None,
is_categorical=None,
alternative=None):
df_prop, df_resp = get_prop_resp(treatment)
df_prop_t10, df_resp_t10 = get_prop_resp("t10a")
metric_values = metric(df_prop)
metric_value = metrics.get_mean(metric_values)
metric_t10_values = metric(df_prop_t10)
metric_value_t10 = metrics.get_mean(metric_t10_values)
metric_values = metrics.get_data(metric_values)
metric_t10_values = metrics.get_data(metric_t10_values)
#print(stats.chisquare(metric_values[:103], metric_t10_values[:103]))
dof = 0
diff = None
print(metric.__name__)
if is_categorical:
#table, res = rp.crosstab(pd.Series(metric_values), pd.Series(metric_t10_values), test='g-test')
table, res = rp.crosstab(pd.Series(metric_values),
pd.Series(metric_t10_values),
test='fisher')
#print(table, res)
#s, p, r = res.results
s = res.results[0]
p = res.results[1]
r = res.results[4]
test_label = f"(g-test chi2)"
print(
"Conclusion: ",
generate_cat_stat_sentence(np.mean(metric_t10_values),
np.std(metric_t10_values),
np.mean(metric_values),
np.std(metric_values),
s,
p,
dof,
diff=diff,
label1="t10a.dss",
label2=treatment + ".dss"))
print(
pd.crosstab(pd.Series(metric_t10_values),
pd.Series(metric_values)))
else:
table, res = rp.ttest(pd.Series(metric_t10_values),
pd.Series(metric_values),
paired=False)
s = res.results[2]
if alternative == "greater":
p = res.results[4]
elif alternative == "less":
p = res.results[5]
elif alternative in (None, 'two-sided'):
p = res.results[3]
r = res.results[9]
diff = res.results[0]
dof = res.results[1]
s = res.results[2]
p = res.results[3]
r = res.results[9]
print(
"Conclusion: ",
generate_stat_sentence(np.mean(metric_t10_values),
np.std(metric_t10_values),
np.mean(metric_values),
np.std(metric_values),
s,
p,
dof,
diff=diff,
label1="t10a.dss",
label2=treatment + ".dss"))
test_label = f"(ttest independent) H0: {'equal' if alternative in {None, 'two-sided'} else alternative}"
print("TABLE: ", table)
print("TEST: ", res)
if as_percentage:
res = {
"Proposer + DSS": f'{100 * metric_value:.2f} %',
"T10": f'{100 * metric_value_t10:.2f} %',
}
else:
res = {
"Proposer + DSS": f'{metric_value:.2f}',
"T10": f'{metric_value_t10:.2f}',
}
if is_categorical:
res[test_label] = f"{s:.3f} (p: {p:.3f}, phi: {r:.3f})"
else:
res[test_label] = f"{s:.3f} (p: {p:.3f}, r: {r:.3f})"
return res