diff --git a/config_files/config.yml b/config_files/config.yml
index 3643419bd75fe149d106a5749b5e37fd190f2616..ff7c3b52f7c5d398c499f6d115f95332d9ab77ed 100644
--- a/config_files/config.yml
+++ b/config_files/config.yml
@@ -52,7 +52,9 @@ algos_multiview: ["all"]
# split, to have more statistically significant results
stats_iter: 1
# The metrics that will be use din the result analysis
-metrics: ["accuracy_score", "f1_score"]
+metrics:
+ "accuracy_score":
+ "f1_score":
# The metric that will be used in the hyper-parameter optimization process
metric_princ: "f1_score"
# The type of hyper-parameter optimization method
diff --git a/config_files/config_test.yml b/config_files/config_test.yml
index af27496b79071a59abdbe0dc7df450d8a35b07e5..2b4c2a2695566f1221546c5ed6ad5cc415c5a166 100644
--- a/config_files/config_test.yml
+++ b/config_files/config_test.yml
@@ -7,8 +7,8 @@ views:
pathf: "examples/data/"
nice: 0
random_state: 42
-nb_cores: 4
-full: False
+nb_cores: 1
+full: True
debug: True
add_noise: False
noise_std: 0.0
@@ -21,7 +21,7 @@ split: 0.8
nb_folds: 2
nb_class: 3
classes:
-type: ["multiview"]
+type: ["monoview", "multiview"]
algos_monoview: ["decision_tree", ]
algos_multiview: ["early_fusion_adaboost"]
stats_iter: 3
@@ -29,7 +29,7 @@ metrics:
accuracy_score: {}
f1_score: {}
metric_princ: "accuracy_score"
-hps_type: "Random"
+hps_type: "None"
hps_args:
n_iter: 10
equivalent_draws: False
@@ -39,6 +39,8 @@ hps_args:
low: 1
high: 10
+decision_tree:
+ max_depth: 1
weighted_linear_early_fusion:
view_weights: null
diff --git a/summit/multiview_platform/exec_classif.py b/summit/multiview_platform/exec_classif.py
index 03ef02a44f3a1a40cf60f2ef835e12b467eb7a97..00aff20e7f1d6828fe3187a3be7e7c646b291965 100644
--- a/summit/multiview_platform/exec_classif.py
+++ b/summit/multiview_platform/exec_classif.py
@@ -465,7 +465,7 @@ def exec_one_benchmark_mono_core(dataset_var=None, labels_dictionary=None,
nb_cores, args["file_type"], args["pathf"], random_state,
>>>>>>> develop
hyper_param_search=hyper_param_search,
- metrics=metrics,
+ metrics=metrics, feature_ids=dataset_var.feature_ids[arguments["view_index"]],
**arguments)]
except BaseException:
if track_tracebacks:
@@ -560,7 +560,9 @@ def exec_benchmark(nb_cores, stats_iter,
analyze_iterations([benchmark_results],
benchmark_arguments_dictionaries, stats_iter,
metrics, sample_ids=dataset_var.sample_ids,
- labels=dataset_var.get_labels())
+ labels=dataset_var.get_labels(),
+ feature_ids=dataset_var.feature_ids,
+ view_names=dataset_var.view_names)
results += [benchmark_results]
logging.info("Done:\t Executing all the needed benchmarks")
@@ -571,7 +573,8 @@ def exec_benchmark(nb_cores, stats_iter,
metrics,
directory,
dataset_var.sample_ids,
- dataset_var.get_labels())
+ dataset_var.get_labels(),dataset_var.feature_ids,
+ dataset_var.view_names)
logging.info("Done:\t Analyzing predictions")
return results_mean_stds
diff --git a/summit/multiview_platform/monoview/exec_classif_mono_view.py b/summit/multiview_platform/monoview/exec_classif_mono_view.py
index 98b65e3dbfde3c8cbbbb51e6096065eb2ef0214e..874a2bfce540665c60fa3623c9e903f24f14e1c7 100644
--- a/summit/multiview_platform/monoview/exec_classif_mono_view.py
+++ b/summit/multiview_platform/monoview/exec_classif_mono_view.py
@@ -54,7 +54,7 @@ def exec_monoview(directory, X, Y, database_name, labels_names,
k_folds, nb_cores, databaseType, path,
random_state, hyper_param_search="Random",
metrics={"accuracy_score*": {}}, n_iter=30, view_name="",
- hps_kwargs={}, **args):
+ hps_kwargs={}, feature_ids=[], **args):
logging.info("Start:\t Loading data")
kwargs, \
t_start, \
@@ -151,7 +151,8 @@ def exec_monoview(directory, X, Y, database_name, labels_names,
labels=Y,
database_name=database_name,
nb_cores=nb_cores,
- duration=whole_duration)
+ duration=whole_duration,
+ feature_ids=feature_ids)
string_analysis, images_analysis, metrics_scores, class_metrics_scores, \
confusion_matrix = result_analyzer.analyze()
logging.info("Done:\t Getting results")
diff --git a/summit/multiview_platform/monoview/monoview_utils.py b/summit/multiview_platform/monoview/monoview_utils.py
index 2fbf53d3e077e013382663ceaa2af5ce57024531..2d227fb93eb1635ef75f1488e286b54e2066989d 100644
--- a/summit/multiview_platform/monoview/monoview_utils.py
+++ b/summit/multiview_platform/monoview/monoview_utils.py
@@ -76,7 +76,20 @@ def gen_test_folds_preds(X_train, y_train, KFolds, estimator):
class BaseMonoviewClassifier(BaseClassifier):
- def get_feature_importance(self, directory, base_file_name,
+ def get_interpretation(self, directory, base_file_name, y_test, feature_ids,
+ multi_class=False):
+ """
+ Base method that returns an empty string if there is not interpretation
+ method in the classifier's module
+ """
+ if hasattr(self, "feature_importances_"):
+ return self.get_feature_importance(directory, base_file_name,
+ feature_ids,)
+ else:
+ return ""
+
+
+ def get_feature_importance(self, directory, base_file_name, feature_ids,
nb_considered_feats=50):
"""Used to generate a graph and a pickle dictionary representing
feature importances"""
@@ -85,28 +98,19 @@ class BaseMonoviewClassifier(BaseClassifier):
feature_importances_sorted = feature_importances[sorted_args][
:nb_considered_feats]
feature_indices_sorted = sorted_args[:nb_considered_feats]
- fig, ax = plt.subplots()
- x = np.arange(len(feature_indices_sorted))
- formatter = FuncFormatter(percent)
- ax.yaxis.set_major_formatter(formatter)
- plt.bar(x, feature_importances_sorted)
- plt.title("Importance depending on feature")
- fig.savefig(
- os.path.join(directory, base_file_name + "feature_importances.png"), transparent=True)
- plt.close()
- features_importances_dict = dict((featureIndex, featureImportance)
- for featureIndex, featureImportance in
+ features_importances_dict = dict((feature_ids[feature_index], feature_importance)
+ for feature_index, feature_importance in
enumerate(feature_importances)
- if featureImportance != 0)
+ if feature_importance != 0)
with open(directory + 'feature_importances.pickle', 'wb') as handle:
pickle.dump(features_importances_dict, handle)
interpret_string = "Feature importances : \n"
- for featureIndex, featureImportance in zip(feature_indices_sorted,
+ for feature_index, feature_importance in zip(feature_indices_sorted,
feature_importances_sorted):
- if featureImportance > 0:
- interpret_string += "- Feature index : " + str(featureIndex) + \
+ if feature_importance > 0:
+ interpret_string += "- Feature : " + feature_ids[feature_index] + \
", feature importance : " + str(
- featureImportance) + "\n"
+ feature_importance) + "\n"
return interpret_string
def get_name_for_fusion(self):
@@ -175,12 +179,12 @@ class MonoviewResultAnalyzer(ResultAnalyser):
classification_indices, k_folds, hps_method, metrics_dict,
n_iter, class_label_names, pred,
directory, base_file_name, labels, database_name, nb_cores,
- duration):
+ duration, feature_ids):
ResultAnalyser.__init__(self, classifier, classification_indices,
k_folds, hps_method, metrics_dict, n_iter,
class_label_names, pred,
directory, base_file_name, labels,
- database_name, nb_cores, duration)
+ database_name, nb_cores, duration, feature_ids)
self.view_name = view_name
self.classifier_name = classifier_name
self.shape = shape
diff --git a/summit/multiview_platform/monoview_classifiers/decision_tree.py b/summit/multiview_platform/monoview_classifiers/decision_tree.py
index 6b309dc3e11f09e165bab1d5c30d9082681de300..33b99090c3149942412296c0bc880d56e30202a6 100644
--- a/summit/multiview_platform/monoview_classifiers/decision_tree.py
+++ b/summit/multiview_platform/monoview_classifiers/decision_tree.py
@@ -34,11 +34,12 @@ class DecisionTree(DecisionTreeClassifier, BaseMonoviewClassifier):
["best", "random"], [random_state]]
self.weird_strings = {}
- def get_interpretation(self, directory, base_file_name, y_test,
+ def get_interpretation(self, directory, base_file_name, y_test, feature_ids,
multiclass=False):
- interpretString = "First featrue : \n\t{} <= {}\n".format(
- self.tree_.feature[0],
+ interpretString = "First feature : \n\t{} <= {}\n".format(
+ feature_ids[self.tree_.feature[0]],
self.tree_.threshold[0])
interpretString += self.get_feature_importance(directory,
- base_file_name)
+ base_file_name,
+ feature_ids)
return interpretString
diff --git a/summit/multiview_platform/multiview/exec_multiview.py b/summit/multiview_platform/multiview/exec_multiview.py
index a594624c8f5aab3e332b6ef7d20776d91308500c..254b11b3a676e0a82e3c7ef6abb6a4dc0a181753 100644
--- a/summit/multiview_platform/multiview/exec_multiview.py
+++ b/summit/multiview_platform/multiview/exec_multiview.py
@@ -61,7 +61,7 @@ def init_constants(kwargs, classification_indices, metrics,
for view_index, view_name in zip(views_indices, views):
logging.info("Info:\t Shape of " + str(view_name) + " :" + str(
- dataset_var.get_shape()))
+ dataset_var.get_shape(view_index)))
labels = dataset_var.get_labels()
directory = os.path.join(directory, classifier_name)
base_file_name = classifier_name + "-" + dataset_var.get_name() + "-"
@@ -266,6 +266,8 @@ def exec_multiview(directory, dataset_var, name, classification_indices,
logging.info("Start:\t Optimizing hyperparameters")
hps_beg = time.monotonic()
+
+ print(dataset_var.view_dict)
if hps_method != "None":
hps_method_class = getattr(hyper_parameter_search, hps_method)
estimator = getattr(classifier_module, classifier_name)(
@@ -294,6 +296,7 @@ def exec_multiview(directory, dataset_var, name, classification_indices,
classifier.fit(dataset_var, dataset_var.get_labels(),
train_indices=learning_indices,
view_indices=views_indices)
+ print("pou")
fit_duration = time.monotonic() - fit_beg
logging.info("Done:\t Fitting classifier")
@@ -332,7 +335,8 @@ def exec_multiview(directory, dataset_var, name, classification_indices,
labels=labels,
database_name=dataset_var.get_name(),
nb_cores=nb_cores,
- duration=whole_duration)
+ duration=whole_duration,
+ feature_ids=dataset_var.feature_ids)
string_analysis, images_analysis, metrics_scores, class_metrics_scores, \
confusion_matrix = result_analyzer.analyze()
logging.info("Done:\t Result Analysis for " + cl_type)
@@ -344,4 +348,4 @@ def exec_multiview(directory, dataset_var, name, classification_indices,
return MultiviewResult(cl_type, classifier_config, metrics_scores,
full_pred, hps_duration, fit_duration,
- pred_duration, class_metrics_scores)
+ pred_duration, class_metrics_scores, classifier)
diff --git a/summit/multiview_platform/multiview/multiview_utils.py b/summit/multiview_platform/multiview/multiview_utils.py
index 6a26aaa37623b1b1f265e10af39ecdc3bef513fa..6f0518a11304a619bf4036278d7136a95ccff7bb 100644
--- a/summit/multiview_platform/multiview/multiview_utils.py
+++ b/summit/multiview_platform/multiview/multiview_utils.py
@@ -133,7 +133,7 @@ def get_monoview_classifier(classifier_name, multiclass=False):
class MultiviewResult(object):
def __init__(self, classifier_name, classifier_config,
metrics_scores, full_labels, hps_duration, fit_duration,
- pred_duration, class_metric_scores):
+ pred_duration, class_metric_scores, clf):
self.classifier_name = classifier_name
self.classifier_config = classifier_config
self.metrics_scores = metrics_scores
@@ -142,6 +142,7 @@ class MultiviewResult(object):
self.fit_duration = fit_duration
self.pred_duration = pred_duration
self.class_metric_scores = class_metric_scores
+ self.clf=clf
def get_classifier_name(self):
try:
@@ -160,7 +161,7 @@ class MultiviewResultAnalyzer(ResultAnalyser):
def __init__(self, view_names, classifier, classification_indices, k_folds,
hps_method, metrics_dict, n_iter, class_label_names,
pred, directory, base_file_name, labels,
- database_name, nb_cores, duration):
+ database_name, nb_cores, duration, feature_ids):
if hps_method.endswith("equiv"):
n_iter = n_iter * len(view_names)
ResultAnalyser.__init__(self, classifier, classification_indices,
@@ -169,7 +170,7 @@ class MultiviewResultAnalyzer(ResultAnalyser):
class_label_names,
pred, directory,
base_file_name, labels, database_name,
- nb_cores, duration)
+ nb_cores, duration, feature_ids)
self.classifier_name = classifier.short_name
self.view_names = view_names
diff --git a/summit/multiview_platform/multiview_classifiers/additions/early_fusion_from_monoview.py b/summit/multiview_platform/multiview_classifiers/additions/early_fusion_from_monoview.py
index 582466b018d351cd3ed88dce8884972bff216e85..8b4f0b01ccf36a48bcf5a91d5beb4191f6a3df9c 100644
--- a/summit/multiview_platform/multiview_classifiers/additions/early_fusion_from_monoview.py
+++ b/summit/multiview_platform/multiview_classifiers/additions/early_fusion_from_monoview.py
@@ -30,8 +30,11 @@ class BaseEarlyFusion(BaseMultiviewClassifier):
return monoview_params
def fit(self, X, y, train_indices=None, view_indices=None):
- train_indices, X = self.transform_data_to_monoview(X, train_indices,
- view_indices)
+ self.view_dict = X.view_dict
+ train_indices, self.view_indices = get_samples_views_indices(X,
+ train_indices,
+ view_indices)
+ train_indices, X = self.transform_data_to_monoview(X, train_indices,)
self.used_views = view_indices
if np.unique(y[train_indices]).shape[0] > 2 and \
not (isinstance(self.monoview_classifier, MultiClassWrapper)):
@@ -40,22 +43,22 @@ class BaseEarlyFusion(BaseMultiviewClassifier):
multiview=False,
y=y[train_indices])
self.monoview_classifier.fit(X, y[train_indices])
+ if hasattr(self.monoview_classifier, "feature_importances_"):
+ self.get_feature_importances()
return self
def predict(self, X, sample_indices=None, view_indices=None):
- _, X = self.transform_data_to_monoview(X, sample_indices, view_indices)
+ _, X = self.transform_data_to_monoview(X, sample_indices)
self._check_views(self.view_indices)
predicted_labels = self.monoview_classifier.predict(X)
return predicted_labels
- def transform_data_to_monoview(self, dataset, sample_indices,
- view_indices):
+ def get_feature_importances(self):
+ self.feature_importances_ = self.monoview_classifier.feature_importances_
+
+ def transform_data_to_monoview(self, dataset, sample_indices):
"""Here, we extract the data from the HDF5 dataset file and store all
the concatenated views in one variable"""
- sample_indices, self.view_indices = get_samples_views_indices(dataset,
- sample_indices,
- view_indices)
-
X = self.hdf5_to_monoview(dataset, sample_indices)
return sample_indices, X
@@ -65,4 +68,8 @@ class BaseEarlyFusion(BaseMultiviewClassifier):
[dataset.get_v(view_idx, samples)
for index, view_idx
in enumerate(self.view_indices)], axis=1)
+ self.feature_ids = []
+ for view_idx in self.view_indices:
+ view_name = dataset.view_names[view_idx]
+ self.feature_ids += [view_name+"-"+feat_id for feat_id in dataset.feature_ids[view_idx]]
return monoview_data
\ No newline at end of file
diff --git a/summit/multiview_platform/result_analysis/error_analysis.py b/summit/multiview_platform/result_analysis/error_analysis.py
index 90d6b69756560a450b304db55abba3cea496f888..aeec9e1e237d5a3e898930c7ff0025b1898c5f5c 100644
--- a/summit/multiview_platform/result_analysis/error_analysis.py
+++ b/summit/multiview_platform/result_analysis/error_analysis.py
@@ -46,7 +46,7 @@ def get_sample_errors(groud_truth, results):
def publish_sample_errors(sample_errors, directory, database_name,
- labels_names, sample_ids, labels): # pragma: no cover
+ label_names, sample_ids, labels): # pragma: no cover
logging.info("Start:\t Label analysis figure generation")
base_file_name = os.path.join(directory, database_name + "-")
@@ -59,7 +59,7 @@ def publish_sample_errors(sample_errors, directory, database_name,
delimiter=",")
plot_2d(data_2d, classifiers_names, nb_classifiers, base_file_name, database_name,
- sample_ids=sample_ids, labels=labels)
+ sample_ids=sample_ids, labels=labels, label_names=label_names)
plot_errors_bar(error_on_samples, nb_samples,
base_file_name, database_name, sample_ids=sample_ids)
@@ -69,7 +69,7 @@ def publish_sample_errors(sample_errors, directory, database_name,
def publish_all_sample_errors(iter_results, directory,
stats_iter,
- sample_ids, labels, data_base_name): # pragma: no cover
+ sample_ids, labels, data_base_name, label_names): # pragma: no cover
logging.info(
"Start:\t Global label analysis figure generation")
@@ -83,7 +83,7 @@ def publish_all_sample_errors(iter_results, directory,
plot_2d(data, classifier_names, nb_classifiers,
os.path.join(directory, ""), data_base_name, stats_iter=stats_iter,
- sample_ids=sample_ids, labels=labels)
+ sample_ids=sample_ids, labels=labels, label_names=label_names)
plot_errors_bar(error_on_samples, nb_samples, os.path.join(directory, ""), data_base_name,
sample_ids=sample_ids)
@@ -152,7 +152,7 @@ def gen_error_data_glob(iter_results, stats_iter):
def plot_2d(data, classifiers_names, nb_classifiers, file_name, dataset_name, labels=None,
- stats_iter=1, use_plotly=True, sample_ids=None): # pragma: no cover
+ stats_iter=1, use_plotly=True, sample_ids=None, label_names=None): # pragma: no cover
r"""Used to generate a 2D plot of the errors.
Parameters
@@ -178,6 +178,8 @@ def plot_2d(data, classifiers_names, nb_classifiers, file_name, dataset_name, la
Returns
-------
"""
+ if label_names is None:
+ label_names = [str(lab) for lab in np.sort(np.unique(labels))]
fig, ax = plt.subplots(nrows=1, ncols=1, )
label_index_list = np.concatenate([np.where(labels == i)[0] for i in
np.unique(
@@ -202,7 +204,7 @@ def plot_2d(data, classifiers_names, nb_classifiers, file_name, dataset_name, la
hover_text = [[sample_ids[sample_index] + " failed " + str(
stats_iter - data[
sample_index, classifier_index]) + " time(s), labelled " + str(
- labels[sample_index])
+ label_names[int(labels[sample_index])])
for classifier_index in range(data.shape[1])]
for sample_index in range(data.shape[0])]
fig = plotly.graph_objs.Figure()
@@ -258,9 +260,11 @@ def plot_errors_bar(error_on_samples, nb_samples, file_name, dataset_name,
fig = plotly.graph_objs.Figure(
[plotly.graph_objs.Bar(x=sample_ids, y=1 - error_on_samples)])
fig.update_layout(paper_bgcolor='rgba(0,0,0,0)',
- plot_bgcolor='rgba(0,0,0,0)')
- title = "Dataset : {} <br> Error % for each sample <br> Generated on <a href='https://baptiste.bauvin.pages.lis-lab.fr/summit'>SuMMIT</a>.".format(
- dataset_name)
+ plot_bgcolor='rgba(0,0,0,0)',
+ title="Dataset : {} <br> Error % for each sample <br> Generated on <a href='https://baptiste.bauvin.pages.lis-lab.fr/summit'>SuMMIT</a>.".format(
+ dataset_name)
+ )
+
plotly.offline.plot(fig, filename=file_name + "error_analysis_bar.html",
auto_open=False)
diff --git a/summit/multiview_platform/result_analysis/execution.py b/summit/multiview_platform/result_analysis/execution.py
index 20ff793b2279430d5b4918c38320d1c6b4861654..279891e7d30ea3fa2262e25eea013a883d61db52 100644
--- a/summit/multiview_platform/result_analysis/execution.py
+++ b/summit/multiview_platform/result_analysis/execution.py
@@ -13,13 +13,14 @@ from .tracebacks_analysis import save_failed, publish_tracebacks
def analyze(results, stats_iter, benchmark_argument_dictionaries,
- metrics, directory, sample_ids, labels): # pragma: no cover
+ metrics, directory, sample_ids, labels, feature_ids,
+ view_names): # pragma: no cover
"""Used to analyze the results of the previous benchmarks"""
data_base_name = benchmark_argument_dictionaries[0]["args"]["name"]
results_means_std, iter_results, flagged_failed, label_names = analyze_iterations(
results, benchmark_argument_dictionaries,
- stats_iter, metrics, sample_ids, labels)
+ stats_iter, metrics, sample_ids, labels, feature_ids, view_names)
if flagged_failed:
save_failed(flagged_failed, directory)
@@ -31,7 +32,7 @@ def analyze(results, stats_iter, benchmark_argument_dictionaries,
def analyze_iterations(results, benchmark_argument_dictionaries, stats_iter,
- metrics, sample_ids, labels):
+ metrics, sample_ids, labels, feature_ids, view_names):
r"""Used to extract and format the results of the different
experimentations performed.
@@ -81,7 +82,9 @@ def analyze_iterations(results, benchmark_argument_dictionaries, stats_iter,
result,
labels_names)
sample_errors = get_sample_errors(labels, result)
- feature_importances = get_feature_importances(result)
+ feature_importances = get_feature_importances(result,
+ feature_ids=feature_ids,
+ view_names=view_names)
durations = get_duration(result)
directory = arguments["directory"]
@@ -124,7 +127,7 @@ def analyze_all(iter_results, stats_iter, directory, data_base_name,
data_base_name, stats_iter,
label_names)
publish_all_sample_errors(error_analysis, directory, stats_iter,
- sample_ids, labels, data_base_name)
+ sample_ids, labels, data_base_name, label_names)
publish_feature_importances(feature_importances, directory,
data_base_name, feature_importances_stds)
plot_durations(duration_means, directory, data_base_name, duration_stds)
diff --git a/summit/multiview_platform/result_analysis/feature_importances.py b/summit/multiview_platform/result_analysis/feature_importances.py
index 86c31366f6b59a8ade8ec547cdd47b6b6ebb1055..e7fec1438bd97aafb1fcabeac4b207cc6cc4fca4 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_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
@@ -28,7 +28,7 @@ def get_feature_importances(result, feature_names=None):
if isinstance(classifier_result, MonoviewResult):
if classifier_result.view_name not in feature_importances:
feature_importances[classifier_result.view_name] = pd.DataFrame(
- index=feature_names)
+ index=feature_ids[classifier_result.view_index])
if hasattr(classifier_result.clf, 'feature_importances_'):
feature_importances[classifier_result.view_name][
classifier_result.classifier_name] = classifier_result.clf.feature_importances_
@@ -36,25 +36,62 @@ def get_feature_importances(result, feature_names=None):
feature_importances[classifier_result.view_name][
classifier_result.classifier_name] = np.zeros(
classifier_result.n_features)
+ else:
+ if "mv" not in feature_importances:
+ feat_ids = []
+ for view_ind, v_feature_id in enumerate(feature_ids):
+ feat_ids += [view_names[view_ind] + "-" + ind for ind in
+ 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_
return feature_importances
def publish_feature_importances(feature_importances, directory, database_name,
feature_stds=None): # pragma: no cover
+ 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 not os.path.exists(os.path.join(directory, "feature_importances")):
- os.mkdir(os.path.join(directory, "feature_importances"))
- file_name = os.path.join(directory, "feature_importances",
- database_name + "-" + view_name
- + "-feature_importances")
+ 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]
+
+
+ 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
+ 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)))))
+ feature_importances_df = pd.concat(importance_dfs[:-1])
+ feature_importances_df = feature_importances_df/feature_importances_df.sum(axis=0)
+ feature_std_df = pd.concat(std_dfs[:-1])
+ 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 = feature_stds[view_name]
- feature_std.to_csv(file_name + "_dataframe_stds.csv")
+ feature_std_df = pd.concat([feature_std_df, feature_stds["mv"]], axis=1,).fillna(0)
else:
- feature_std = pd.DataFrame(data=np.zeros(feature_importance.shape),
- index=feature_importance.index,
- columns=feature_importance.columns)
- plot_feature_importances(file_name, feature_importance, feature_std)
+ 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)
def plot_feature_importances(file_name, feature_importance,
diff --git a/summit/multiview_platform/utils/base.py b/summit/multiview_platform/utils/base.py
index c88294c59977b3160ddada9b1a343f8fa8f12909..67df47a6eb850da32faa5ea132b1e73a0d746ace 100644
--- a/summit/multiview_platform/utils/base.py
+++ b/summit/multiview_platform/utils/base.py
@@ -105,7 +105,7 @@ class BaseClassifier(BaseEstimator, ):
else:
return str(self.get_params()[param_name])
- def get_interpretation(self, directory, base_file_name, y_test,
+ def get_interpretation(self, directory, base_file_name, y_test, feature_ids,
multi_class=False):
"""
Base method that returns an empty string if there is not interpretation
@@ -169,7 +169,7 @@ class ResultAnalyser():
def __init__(self, classifier, classification_indices, k_folds,
hps_method, metrics_dict, n_iter, class_label_names,
pred, directory, base_file_name, labels,
- database_name, nb_cores, duration):
+ database_name, nb_cores, duration, feature_ids):
"""
Parameters
@@ -219,6 +219,7 @@ class ResultAnalyser():
self.duration = duration
self.metric_scores = {}
self.class_metric_scores = {}
+ self.feature_ids=feature_ids
def get_all_metrics_scores(self, ):
"""
@@ -387,7 +388,7 @@ class ResultAnalyser():
string_analysis += "\n\n Classifier Interpretation : \n"
string_analysis += self.classifier.get_interpretation(
self.directory, self.base_file_name,
- self.labels[self.test_indices])
+ self.labels[self.test_indices], self.feature_ids)
image_analysis = {}
return string_analysis, image_analysis, self.metric_scores, \
self.class_metric_scores, self.confusion_matrix
diff --git a/summit/multiview_platform/utils/dataset.py b/summit/multiview_platform/utils/dataset.py
index 07023b756fea909cc01e821de05f1f28febf61b8..2a33b34bac58766b39049efa68614ddde26b9522 100644
--- a/summit/multiview_platform/utils/dataset.py
+++ b/summit/multiview_platform/utils/dataset.py
@@ -165,12 +165,18 @@ class Dataset():
return selected_label_names
+ def gen_feat_id(self):
+ self.feature_ids = [["ID_" + str(i) for i in
+ range(self.get_v(view_ind).shape[1])]
+ for view_ind in self.view_dict.values()]
+
+
class RAMDataset(Dataset):
def __init__(self, views=None, labels=None, are_sparse=False,
view_names=None, labels_names=None, sample_ids=None,
- name=None):
+ name=None, feature_ids=None):
self.saved_on_disk = False
self.views = views
self.labels = np.asarray(labels)
@@ -187,6 +193,13 @@ class RAMDataset(Dataset):
self.name = name
self.nb_view = len(self.views)
self.is_temp = False
+ if feature_ids is not None:
+ feature_ids = [[feature_id if not is_just_number(feature_id)
+ else "ID_" + feature_id for feature_id in
+ feat_ids] for feat_ids in feature_ids]
+ self.feature_ids = feature_ids
+ else:
+ self.gen_feat_id()
def get_view_name(self, view_idx):
return self.view_names[view_idx]
@@ -319,7 +332,7 @@ class HDF5Dataset(Dataset):
def __init__(self, views=None, labels=None, are_sparse=False,
file_name="dataset.hdf5", view_names=None, path="",
hdf5_file=None, labels_names=None, is_temp=False,
- sample_ids=None, ):
+ sample_ids=None, feature_ids=None):
self.is_temp = False
if hdf5_file is not None:
self.dataset = hdf5_file
@@ -364,6 +377,14 @@ class HDF5Dataset(Dataset):
else:
self.sample_ids = ["ID_" + str(i)
for i in range(labels.shape[0])]
+ if feature_ids is not None:
+ feature_ids = [[feature_id if not is_just_number(feature_id)
+ else "ID_" + feature_id for feature_id in
+ feat_ids] for feat_ids in feature_ids]
+ self.feature_ids = feature_ids
+ else:
+ self.gen_feat_id()
+
def get_v(self, view_index, sample_indices=None):
""" Extract the view and returns a numpy.ndarray containing the description
@@ -423,6 +444,7 @@ class HDF5Dataset(Dataset):
"""
self.nb_view = self.dataset["Metadata"].attrs["nbView"]
self.view_dict = self.get_view_dict()
+ self.view_names = [self.dataset["View{}".format(ind)].attrs['name'] for ind in range(self.nb_view)]
if "sample_ids" in self.dataset["Metadata"].keys():
self.sample_ids = [sample_id.decode()
if not is_just_number(sample_id.decode())
@@ -432,6 +454,14 @@ class HDF5Dataset(Dataset):
else:
self.sample_ids = ["ID_" + str(i) for i in
range(self.dataset["Labels"].shape[0])]
+ if "feature_ids" in self.dataset["Metadata"].keys():
+ self.feature_ids = [[feature_id.decode()
+ if not is_just_number(feature_id.decode())
+ else "ID_" + feature_id.decode()
+ for feature_id in feature_ids] for feature_ids in
+ self.dataset["Metadata"]["feature_ids"]]
+ else:
+ self.gen_feat_id()
def get_nb_samples(self):
"""