diff --git a/summit/multiview_platform/result_analysis/execution.py b/summit/multiview_platform/result_analysis/execution.py
index 931d61863f3ae12128ce2c2c8c4933312045ff00..b503e007c800a735c5eaed0b8fc03f7dbc8b358a 100644
--- a/summit/multiview_platform/result_analysis/execution.py
+++ b/summit/multiview_platform/result_analysis/execution.py
@@ -84,7 +84,7 @@ def analyze_iterations(results, benchmark_argument_dictionaries, stats_iter,
sample_errors = get_sample_errors(labels, result)
feature_importances = get_feature_importances(result,
feature_ids=feature_ids,
- view_names=view_names,)
+ view_names=view_names, )
durations = get_duration(result)
directory = arguments["directory"]
diff --git a/summit/multiview_platform/result_analysis/feature_importances.py b/summit/multiview_platform/result_analysis/feature_importances.py
index 36c0eb3514b0fa3db388af10803b60f2f245f011..1c6b4188035f3a16c0d7e246091458db2816c88b 100644
--- a/summit/multiview_platform/result_analysis/feature_importances.py
+++ b/summit/multiview_platform/result_analysis/feature_importances.py
@@ -7,7 +7,7 @@ import plotly
from ..monoview.monoview_utils import MonoviewResult
-def get_feature_importances(result, feature_ids=None, view_names=None,):
+def get_feature_importances(result, feature_ids=None, view_names=None, ):
r"""Extracts the feature importance from the monoview results and stores
them in a dictionnary :
feature_importance[view_name] is a pandas.DataFrame of size n_feature*n_clf
@@ -44,75 +44,67 @@ def get_feature_importances(result, feature_ids=None, view_names=None,):
v_feature_id]
feature_importances["mv"] = pd.DataFrame(index=feat_ids)
if hasattr(classifier_result.clf, 'feature_importances_'):
- feature_importances["mv"][classifier_result.classifier_name] = classifier_result.clf.feature_importances_
+ feature_importances["mv"][classifier_result.get_classifier_name()] = classifier_result.clf.feature_importances_
+ else:
+ # HACK: Assigning a default features importances values to classifier that hasn't feature_importances_
+ # attribute (eg: Linear Late Fusion)
+ feature_importances["mv"][classifier_result.get_classifier_name()] = np.zeros(len(feature_importances["mv"].index))
return feature_importances
def publish_feature_importances(feature_importances, directory, database_name,
feature_stds=None, metric_scores=None): # pragma: no cover
+ # TODO: Manage the case with NAN values
importance_dfs = []
std_dfs = []
if not os.path.exists(os.path.join(directory, "feature_importances")):
os.mkdir(os.path.join(directory, "feature_importances"))
for view_name, feature_importance in feature_importances.items():
- if view_name!="mv":
+ if feature_stds is not None:
+ feature_std = feature_stds[view_name]
+ else:
+ feature_std = pd.DataFrame(data=np.zeros(feature_importance.shape),
+ index=feature_importance.index,
+ columns=feature_importance.columns)
+ feature_std = feature_std.loc[feature_importance.index]
+
+ if view_name == "mv":
+ importance_dfs.append(feature_importance)
+ std_dfs.append(feature_std)
+ else:
+ importance_dfs.append(feature_importance.set_index(
+ pd.Index([view_name + "-" + ind for ind in list(feature_importance.index)])))
- if feature_stds is not None:
- feature_std = feature_stds[view_name]
- else:
- feature_std = pd.DataFrame(data=np.zeros(feature_importance.shape),
- index=feature_importance.index,
- columns=feature_importance.columns)
- feature_std = feature_std.loc[feature_importance.index]
-
-
- importance_dfs.append(feature_importance.set_index(pd.Index([view_name+"-"+ind for ind in list(feature_importance.index)])))
- # importance_dfs.append(pd.DataFrame(index=[view_name+"-br"],
- # columns=feature_importance.columns,
- # data=np.zeros((1, len(
- # feature_importance.columns)))))
- std_dfs.append(feature_std.set_index(pd.Index([view_name+"-"+ind
+ std_dfs.append(feature_std.set_index(pd.Index([view_name + "-" + ind
for ind
in list(feature_std.index)])))
- # std_dfs.append(pd.DataFrame(index=[view_name + "-br"],
- # columns=feature_std.columns,
- # data=np.zeros((1, len(
- # feature_std.columns)))))
- if len(importance_dfs)>0:
+
+ if len(importance_dfs) > 0:
feature_importances_df = pd.concat(importance_dfs)
- feature_importances_df = feature_importances_df/feature_importances_df.sum(axis=0)
+ feature_importances_df = feature_importances_df / feature_importances_df.sum(axis=0)
feature_std_df = pd.concat(std_dfs)
- if "mv" in feature_importances:
- feature_importances_df = pd.concat([feature_importances_df,feature_importances["mv"].loc[(feature_importances["mv"] != 0).any(axis=1), :]], axis=1).fillna(0)
- if feature_stds is not None:
- feature_std_df = pd.concat([feature_std_df, feature_stds["mv"]], axis=1,).fillna(0)
- else:
- fake = pd.DataFrame(data=np.zeros((feature_importances_df.shape[0], feature_importances["mv"].shape[1])),
- index=feature_importances_df.index,
- columns=feature_importances["mv"].columns).fillna(0)
- feature_std_df = pd.concat([feature_std_df, fake], axis=1,).fillna(0)
plot_feature_importances(os.path.join(directory, "feature_importances",
- database_name), feature_importances_df, feature_std_df)
+ database_name), feature_importances_df, feature_std_df)
if metric_scores is not None:
plot_feature_relevance(os.path.join(directory, "feature_importances",
- database_name), feature_importances_df, feature_std_df, metric_scores)
+ database_name), feature_importances_df, feature_std_df, metric_scores)
def plot_feature_importances(file_name, feature_importance,
feature_std): # pragma: no cover
s = feature_importance.sum(axis=1)
- s = s[s!=0]
+ s = s[s != 0]
feature_importance = feature_importance.loc[s.sort_values(ascending=False).index]
feature_importance.to_csv(file_name + "_dataframe.csv")
hover_text = [["-Feature :" + str(feature_name) +
"<br>-Classifier : " + classifier_name +
"<br>-Importance : " + str(
feature_importance.loc[feature_name][classifier_name]) +
- "<br>-STD : " + str(
+ "<br>-STD : " + str(
feature_std.loc[feature_name][classifier_name])
- for classifier_name in list(feature_importance.columns)]
- for feature_name in list(feature_importance.index)]
+ for classifier_name in list(feature_importance.columns)]
+ for feature_name in list(feature_importance.index)]
fig = plotly.graph_objs.Figure(data=plotly.graph_objs.Heatmap(
x=list(feature_importance.columns),
y=list(feature_importance.index),
@@ -129,20 +121,20 @@ def plot_feature_importances(file_name, feature_importance,
plotly.offline.plot(fig, filename=file_name + ".html", auto_open=False)
del fig
-
+
def plot_feature_relevance(file_name, feature_importance,
- feature_std, metric_scores): # pragma: no cover
+ feature_std, metric_scores): # pragma: no cover
for metric, score_df in metric_scores.items():
if metric.endswith("*"):
if isinstance(score_df, dict):
score_df = score_df["mean"]
for score in score_df.columns:
- if len(score.split("-"))>1:
+ if len(score.split("-")) > 1:
algo, view = score.split("-")
feature_importance[algo].loc[[ind for ind in feature_importance.index if ind.startswith(view)]]*=score_df[score]['test']
else:
feature_importance[score] *= score_df[score]['test']
- file_name+="_relevance"
+ file_name += "_relevance"
plot_feature_importances(file_name, feature_importance,
feature_std)