def _execute(self):
df = dicts_to_df(self.features)
format_feature_df(df)
df_real = df[df["Model"] == "real-world"]
print(collections.Counter(df["Model"]))
small_avg_degree = df_real[
"Centrality.Degree.Location.Arithmetic Mean"] <= 30
filters = {
"all": [True] * len(df_real),
"avg-degree-le-30": small_avg_degree,
"avg-degree-gt-30": ~small_avg_degree,
"socfb": df_real["Type"] == "socfb",
"not-socfb": df_real["Type"] != "socfb"
}
format_str = "{:20}{:>5}"
network_models = sorted(
set(
filter(lambda model: not model.endswith("-second"),
set(df["Model"]))) - set(["real-world"]))
for filtername, filterdf in sorted(filters.items()):
graphs = sorted(df_real[filterdf]["Graph"])
print(format_str.format(filtername, len(graphs)))
features_collection = get_all_feature_sets_self_check(df, graphs)
sub_df = df.loc(axis=0)[:, graphs, :]
accuracies = \
classification_experiment(
sub_df,
network_models,
features_collection,
self.cores)
accuracies.to_csv(self._stagepath + "accuracies/" + filtername +
".csv",
header=True,
index_label="features")
def _execute(self):
df = dicts_to_df(self.features)
df.sort_index(axis=1, inplace=True)
format_feature_df(df)
network_models = sorted(set(filter(lambda model: not model.endswith("-second"), set(df["Model"])))-set(["real-world"]))
diff_features = df.columns[df.dtypes == float].values
print("Calculating difference for {} features for {} models...".format(len(diff_features), len(network_models)))
idx = pandas.IndexSlice
for model in network_models:
val_1 = df.loc[idx[:,:,model],diff_features].values
val_2 = df.loc[idx[:,:,model+"-second"],diff_features].values
df.loc[idx[:,:,model],diff_features] = val_1 - val_2
df.loc[idx[:,:,model+"-second"],diff_features] = val_2 - val_1
print("Done with model {}".format(model))
for a_dict in df.to_dict("records"):
self._save_as_csv(a_dict)