def get_explanation_error_df():
kwargs = {
"nsamples": 1000,
"background": "linear_consecutive",
"pen": 1,
"model": "rbf",
"jump": 5,
"plot": False,
"figsize": (20, 3),
"segments_size": None
}
lasts_synth_pat = load_multiple_lasts(
"./saved/pat/pat/synth/matched_uniform_saxdt", simple_load=True)
lasts_synth_pat0 = load_multiple_lasts(
"./saved/pat/pat0/synth/matched_uniform_saxdt", simple_load=True)
lasts_synth_pat1 = load_multiple_lasts(
"./saved/pat/pat1/synth/matched_uniform_saxdt", simple_load=True)
lasts_synth_pat2 = load_multiple_lasts(
"./saved/pat/pat2/synth/matched_uniform_saxdt", simple_load=True)
lasts_synth_pat3 = load_multiple_lasts(
"./saved/pat/pat3/synth/matched_uniform_saxdt", simple_load=True)
lasts_lists = [
lasts_synth_pat, lasts_synth_pat0, lasts_synth_pat1, lasts_synth_pat2,
lasts_synth_pat3
]
names = [
"lasts_synth_pat", "lasts_synth_pat0", "lasts_synth_pat1",
"lasts_synth_pat2", "lasts_synth_pat3"
]
df = defaultdict(list)
for j, lasts_list in enumerate(lasts_lists):
for i, lasts_ in enumerate(lasts_list):
print(i + 1, "/", len(lasts_list))
# df["lasts_" + names[j]].append(lasts_.explanation_error(divide_by_baseline=False))
shapts = ShapTimeSeries()
shapts.shap_values(lasts_.x, lasts_.blackbox, **kwargs)
df["shap_abs_" + names[j]].append(
shapts.explanation_error(divide_by_baseline=False))
print(pd.DataFrame(df))
return pd.DataFrame(df)
def get_predictions_synth():
lasts_synth_pat = load_multiple_lasts(
"./saved/pat/pat/synth/matched_uniform_saxdt", simple_load=True)
lasts_synth_pat0 = load_multiple_lasts(
"./saved/pat/pat0/synth/matched_uniform_saxdt", simple_load=True)
lasts_synth_pat1 = load_multiple_lasts(
"./saved/pat/pat1/synth/matched_uniform_saxdt", simple_load=True)
lasts_synth_pat2 = load_multiple_lasts(
"./saved/pat/pat2/synth/matched_uniform_saxdt", simple_load=True)
lasts_synth_pat3 = load_multiple_lasts(
"./saved/pat/pat3/synth/matched_uniform_saxdt", simple_load=True)
lasts_lists = [
lasts_synth_pat, lasts_synth_pat0, lasts_synth_pat1, lasts_synth_pat2,
lasts_synth_pat3
]
names = [
"lasts_synth_pat", "lasts_synth_pat0", "lasts_synth_pat1",
"lasts_synth_pat2", "lasts_synth_pat3"
]
dict_exp = defaultdict(list)
dict_true = defaultdict(list)
dict_tss = defaultdict(list)
i = 0
for lasts_list, name in zip(lasts_lists, names):
print(i + 1)
for lasts_ in lasts_list:
pred_importances, tss = lasts_.surrogate.predict_explanation(
lasts_.z_tilde, 1)
true_importances = lasts_.blackbox.predict_explanation(
lasts_.z_tilde)
dict_tss[name].append(tss)
dict_true[name].append(true_importances)
dict_exp[name].append(pred_importances)
dump(dict_exp, "./stab_temp/dict_exp_" + str(i) + ".joblib")
dump(dict_true, "./stab_temp/dict_true_" + str(i) + ".joblib")
dump(dict_tss, "./stab_temp/dict_tss_" + str(i) + ".joblib")
i += 1
return dict_exp, dict_true, dict_tss
def get_usefulness_df():
cbf_X_exp = build_cbf(n_samples=600, random_state=0, verbose=False)[10:]
ecg200_X_exp = build_ecg200(verbose=False, random_state=0)[10:]
coffee_X_exp = build_coffee(verbose=False, random_state=0)[10:]
gunpoint_X_exp = build_gunpoint(verbose=False, random_state=0)[10:]
lasts_cbf_knn = load_multiple_lasts(
"./saved/knn/cbf/matched_uniform_saxdt", simple_load=True)
lasts_cbf_resnet = load_multiple_lasts(
"./saved/resnet/cbf/matched_uniform_saxdt", simple_load=True)
lasts_coffee_knn = load_multiple_lasts(
"./saved/knn/coffee/matched_uniform_saxdt", simple_load=True)
lasts_coffee_resnet = load_multiple_lasts(
"./saved/resnet/coffee/matched_uniform_saxdt", simple_load=True)
lasts_ecg200_knn = load_multiple_lasts(
"./saved/knn/ecg200/matched_uniform_saxdt", simple_load=True)
lasts_ecg200_resnet = load_multiple_lasts(
"./saved/resnet/ecg200/matched_uniform_saxdt", simple_load=True)
lasts_gunpoint_knn = load_multiple_lasts(
"./saved/knn/gunpoint/matched_uniform_saxdt", simple_load=True)
lasts_gunpoint_resnet = load_multiple_lasts(
"./saved/resnet/gunpoint/matched_uniform_saxdt", simple_load=True)
dataset_list = [cbf_X_exp, ecg200_X_exp, coffee_X_exp, gunpoint_X_exp]
lasts_lists = [
lasts_cbf_knn, lasts_cbf_resnet, lasts_coffee_knn, lasts_coffee_resnet,
lasts_ecg200_knn, lasts_ecg200_resnet, lasts_gunpoint_knn,
lasts_gunpoint_resnet
]
lasts_lists_names = [
"lasts_cbf_knn", "lasts_cbf_resnet", "lasts_coffee_knn",
"lasts_coffee_resnet", "lasts_ecg200_knn", "lasts_ecg200_resnet",
"lasts_gunpoint_knn", "lasts_gunpoint_resnet"
]
global_names = [
"global_cbf", "global_coffee", "global_ecg200", "global_gunpoint"
]
df_dict = dict()
for dataset, name in zip(dataset_list, global_names):
df_dict[name] = usefulness_scores_real(dataset[0], dataset[1])
for lasts_list, name in zip(lasts_lists, lasts_lists_names):
df_dict[name] = usefulness_scores_lasts(lasts_list)
return df_dict
def get_predictions():
lasts_cbf_knn = load_multiple_lasts(
"./saved/knn/cbf/matched_uniform_saxdt", simple_load=True)
lasts_cbf_resnet = load_multiple_lasts(
"./saved/resnet/cbf/matched_uniform_saxdt", simple_load=True)
lasts_coffee_knn = load_multiple_lasts(
"./saved/knn/coffee/matched_uniform_saxdt", simple_load=True)
lasts_coffee_resnet = load_multiple_lasts(
"./saved/resnet/coffee/matched_uniform_saxdt", simple_load=True)
lasts_ecg200_knn = load_multiple_lasts(
"./saved/knn/ecg200/matched_uniform_saxdt", simple_load=True)
lasts_ecg200_resnet = load_multiple_lasts(
"./saved/resnet/ecg200/matched_uniform_saxdt", simple_load=True)
lasts_gunpoint_knn = load_multiple_lasts(
"./saved/knn/gunpoint/matched_uniform_saxdt", simple_load=True)
lasts_gunpoint_resnet = load_multiple_lasts(
"./saved/resnet/gunpoint/matched_uniform_saxdt", simple_load=True)
lasts_lists = [
lasts_cbf_knn, lasts_cbf_resnet, lasts_coffee_knn, lasts_coffee_resnet,
lasts_ecg200_knn, lasts_ecg200_knn, lasts_ecg200_resnet,
lasts_gunpoint_knn, lasts_gunpoint_resnet
]
names = [
"cbf_knn", "cbf_resnet", "coffee_knn", "coffee_resnet", "ecg200_knn",
"ecg200_knn", "ecg200_resnet", "gunpoint_knn", "gunpoint_resnet"
]
dict_exp = defaultdict(list)
dict_tss = defaultdict(list)
i = 0
for lasts_list, name in zip(lasts_lists, names):
print(i + 1)
for lasts_ in lasts_list:
pred_importances, tss = lasts_.surrogate.predict_explanation(
lasts_.z_tilde, 1)
dict_tss[name].append(tss)
dict_exp[name].append(pred_importances)
dump(dict_exp, "./stab_temp/dict_exp_" + str(i) + ".joblib")
dump(dict_tss, "./stab_temp/dict_tss_" + str(i) + ".joblib")
i += 1
return dict_exp, dict_tss
def get_instability_df(latent=False, compute_lasts=True, compute_shap=True):
shap_kwargs = {
"nsamples": 1000,
"background": "linear_consecutive",
"pen": 1,
"model": "rbf",
"jump": 5,
"plot": False,
"figsize": (20, 3),
"segments_size": None
}
lasts_cbf_knn = load_multiple_lasts(
"./saved/knn/cbf/matched_uniform_saxdt", simple_load=True)
lasts_cbf_resnet = load_multiple_lasts(
"./saved/resnet/cbf/matched_uniform_saxdt", simple_load=True)
lasts_coffee_knn = load_multiple_lasts(
"./saved/knn/coffee/matched_uniform_saxdt", simple_load=True)
lasts_coffee_resnet = load_multiple_lasts(
"./saved/resnet/coffee/matched_uniform_saxdt", simple_load=True)
lasts_ecg200_knn = load_multiple_lasts(
"./saved/knn/ecg200/matched_uniform_saxdt", simple_load=True)
lasts_ecg200_resnet = load_multiple_lasts(
"./saved/resnet/ecg200/matched_uniform_saxdt", simple_load=True)
lasts_gunpoint_knn = load_multiple_lasts(
"./saved/knn/gunpoint/matched_uniform_saxdt", simple_load=True)
lasts_gunpoint_resnet = load_multiple_lasts(
"./saved/resnet/gunpoint/matched_uniform_saxdt", simple_load=True)
lasts_lists = [
lasts_cbf_knn, lasts_cbf_resnet, lasts_coffee_knn, lasts_coffee_resnet,
lasts_ecg200_knn, lasts_ecg200_knn, lasts_ecg200_resnet,
lasts_gunpoint_knn, lasts_gunpoint_resnet
]
names = [
"cbf_knn", "cbf_resnet", "coffee_knn", "coffee_resnet", "ecg200_knn",
"ecg200_knn", "ecg200_resnet", "gunpoint_knn", "gunpoint_resnet"
]
df_dict = dict()
i = 0
for lasts_list, name in zip(lasts_lists, names):
print(i + 1, "/", len(lasts_lists))
i += 1
df1 = instability_multiple(lasts_list=lasts_list,
combination_name=name,
shap_kwargs=shap_kwargs,
latent=True,
compute_lasts=compute_lasts,
compute_shap=compute_shap)
df2 = instability_multiple(lasts_list=lasts_list,
combination_name=name,
shap_kwargs=shap_kwargs,
latent=False,
compute_lasts=compute_lasts,
compute_shap=compute_shap)
df_dict[name + "_lasts_latent"] = df1
df_dict[name + "_lasts_manifest"] = df2
dump(df_dict, "./stability_saved/df_dict_" + str(i - 1) + ".joblib")
return df_dict