nkululeko 0.89.1__tar.gz → 0.90.0__tar.gz

{nkululeko-0.89.1 → nkululeko-0.90.0}/CHANGELOG.md

@@ -1,6 +1,16 @@
 Changelog
 =========
 
+Version 0.90.0
+--------------
+* augmentation can now be done without target
+* random splicing params configurable
+* made kde default for plot continous/categorical plots
+
+Version 0.89.2
+--------------
+* fix shap value calculation
+  
 Version 0.89.1
 --------------
 * print and save result of feature importance

{nkululeko-0.89.1/nkululeko.egg-info → nkululeko-0.90.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.89.1
+Version: 0.90.0
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -356,6 +356,16 @@ F. Burkhardt, Johannes Wagner, Hagen Wierstorf, Florian Eyben and Björn Schulle
 Changelog
 =========
 
+Version 0.90.0
+--------------
+* augmentation can now be done without target
+* random splicing params configurable
+* made kde default for plot continous/categorical plots
+
+Version 0.89.2
+--------------
+* fix shap value calculation
+  
 Version 0.89.1
 --------------
 * print and save result of feature importance

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/augment.py

@@ -37,8 +37,8 @@ def doit(config_file):
 
     filename = util.config_val("AUGMENT", "result", "augmented.csv")
     filename = f"{expr.data_dir}/{filename}"
-
-    if os.path.exists(filename):
+    no_reuse = eval(util.config_val("DATA", "no_reuse", "False"))
+    if os.path.exists(filename) and not no_reuse:
         util.debug("files already augmented")
     else:
         # load the data

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/augmenting/randomsplicer.py

@@ -41,14 +41,17 @@ class Randomsplicer:
             * top_db: top db level for silence to be recognized (default: 12)
         """
 
-        p_reverse = 0.3
-        top_db = 12
+        p_reverse = float(self.util.config_val("AUGMENT", "p_reverse", "0.3"))
+        top_db = float(self.util.config_val("AUGMENT", "top_db", "12"))
 
         files = self.df.index.get_level_values(0).values
         store = self.util.get_path("store")
         filepath = f"{store}randomspliced/"
         audeer.mkdir(filepath)
-        self.util.debug(f"random splicing {sample_selection} samples to {filepath}")
+        self.util.debug(
+            f"random splicing {sample_selection} samples to {filepath}, "
+            + f"p_reverse = {p_reverse}, top_db = {top_db}",
+        )
         newpath = ""
         index_map = {}
         for i, f in enumerate(tqdm(files)):

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/constants.py

@@ -1,2 +1,2 @@
-VERSION="0.89.1"
+VERSION="0.90.0"
 SAMPLING_RATE = 16000

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/data/dataset.py

@@ -30,8 +30,8 @@ class Dataset:
     def __init__(self, name):
         """Constructor setting up name and configuration"""
         self.name = name
-        self.target = glob_conf.config["DATA"]["target"]
         self.util = Util("dataset")
+        self.target = self.util.config_val("DATA", "target", "none")
         self.plot = Plots()
         self.limit = int(self.util.config_val_data(self.name, "limit", 0))
         self.start_fresh = eval(self.util.config_val("DATA", "no_reuse", "False"))
@@ -127,6 +127,9 @@ class Dataset:
             self.got_gender,
             self.got_age,
         ) = self._get_df_for_lists(self.db, df_files_tables)
+        if df.shape[0] > 0 and self.target == "none":
+            self.df = df
+            return
         if False in {
             self.is_labeled,
             self.got_speaker,
@@ -553,7 +556,10 @@ class Dataset:
             " samples in train/test"
         )
         # because this generates new train/test sample quantaties, the feature extraction has to be done again
-        glob_conf.config["FEATS"]["needs_feature_extraction"] = "True"
+        try:
+            glob_conf.config["FEATS"]["needs_feature_extraction"] = "True"
+        except KeyError:
+            pass
 
     def random_split(self):
         """One way to split train and eval sets: Specify percentage of random samples"""

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/ensemble.py

@@ -18,6 +18,7 @@ Raises:
 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 
+
 from typing import List
 import configparser
 import time
@@ -26,10 +27,15 @@ from pathlib import Path
 
 import numpy as np
 import pandas as pd
+import matplotlib.pyplot as plt
+
 from sklearn.metrics import(
+    RocCurveDisplay,
     balanced_accuracy_score, 
     classification_report,
-    f1_score
+    auc,
+    roc_auc_score,
+    roc_curve
 )
 
 from nkululeko.constants import VERSION
@@ -289,9 +295,7 @@ def ensemble_predictions(
     uar = balanced_accuracy_score(truth, predicted)
     acc = (truth == predicted).mean()
     # print classification report
-    Util("ensemble").debug(f"\n {classification_report(truth, predicted)}")
-    # f1 = f1_score(truth, predicted, pos_label='p')
-    # Util("ensemble").debug(f"F1: {f1:.3f}")
+    Util("ensemble").debug(f"\n {classification_report(truth, predicted, digits=4)}")
     Util("ensemble").debug(f"{method}: UAR: {uar:.3f}, ACC: {acc:.3f}")
 
     return ensemble_preds

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/experiment.py

@@ -101,12 +101,15 @@ class Experiment:
             if data.got_speaker:
                 self.got_speaker = True
             self.datasets.update({d: data})
-        self.target = self.util.config_val("DATA", "target", "emotion")
+        self.target = self.util.config_val("DATA", "target", "none")
         glob_conf.set_target(self.target)
         # print target via debug
         self.util.debug(f"target: {self.target}")
         # print keys/column
         dbs = ",".join(list(self.datasets.keys()))
+        if self.target == "none":
+            self.util.debug(f"loaded databases {dbs}")
+            return
         labels = self.util.config_val("DATA", "labels", False)
         auto_labels = list(next(iter(self.datasets.values())).df[self.target].unique())
         if labels:
@@ -191,7 +194,8 @@ class Experiment:
             self.df_train, self.df_test = pd.DataFrame(), pd.DataFrame()
             for d in self.datasets.values():
                 d.split()
-                d.prepare_labels()
+                if self.target != "none":
+                    d.prepare_labels()
                 if d.df_train.shape[0] == 0:
                     self.util.debug(f"warn: {d.name} train empty")
                 self.df_train = pd.concat([self.df_train, d.df_train])
@@ -207,6 +211,8 @@ class Experiment:
             self.df_test.to_csv(storage_test)
             self.df_train.to_csv(storage_train)
 
+        if self.target == "none":
+            return
         self.util.copy_flags(self, self.df_test)
         self.util.copy_flags(self, self.df_train)
         # Try data checks

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/explore.py

@@ -91,7 +91,9 @@ def main(src_dir):
             # these investigations need features to explore
             expr.extract_feats()
             needs_feats = True
-        # explore
+            # explore
+            expr.init_runmanager()
+            expr.runmgr.do_runs()
     expr.analyse_features(needs_feats)
     expr.store_report()
     print("DONE")

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/feat_extract/feats_analyser.py

@@ -50,19 +50,32 @@ class FeatureAnalyser:
 
         name = "my_shap_values"
         if not self.util.exist_pickle(name):
-
+            # get model name
+            model_name = self.util.get_model_type()
+            if hasattr(model, "predict_shap"):
+                model_func = model.predict_shap
+            elif hasattr(model, "clf"):
+                model_func = model.clf.predict
+            else:
+                raise Exception("Model not supported for SHAP analysis")
+            
+            self.util.debug(f"using SHAP explainer for {model_name} model")
+            
             explainer = shap.Explainer(
-                model.predict_shap,
+                model_func,
                 self.features,
                 output_names=glob_conf.labels,
                 algorithm="permutation",
                 npermutations=5,
             )
+ 
             self.util.debug("computing SHAP values...")
             shap_values = explainer(self.features)
             self.util.to_pickle(shap_values, name)
         else:
             shap_values = self.util.from_pickle(name)
+        # plt.figure()
+        plt.close('all')
         plt.tight_layout()
         shap.plots.bar(shap_values)
         fig_dir = self.util.get_path("fig_dir") + "../"  # one up because of the runs
@@ -71,7 +84,8 @@ class FeatureAnalyser:
         filename = f"_SHAP_{model.name}"
         filename = f"{fig_dir}{exp_name}{filename}.{format}"
         plt.savefig(filename)
-        self.util.debug(f"plotted SHAP feature importance tp {filename}")
+        plt.close()
+        self.util.debug(f"plotted SHAP feature importance to {filename}")
 
     def analyse(self):
         models = ast.literal_eval(self.util.config_val("EXPL", "model", "['log_reg']"))

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/modelrunner.py

@@ -53,8 +53,8 @@ class Modelrunner:
             # epochs are handled by Huggingface API
             self.model.train()
             report = self.model.predict()
-            # todo: findout the best epoch, no need
-            # since oad_best_model_at_end is given in training args
+            # todo: findout the best epoch -> no need
+            # since load_best_model_at_end is given in training args
             epoch = epoch_num
             report.set_id(self.run, epoch)
             plot_name = self.util.get_plot_name() + f"_{self.run}_{epoch:03d}_cnf"

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/plots.py

@@ -263,7 +263,7 @@ class Plots:
 
     def plotcatcont(self, df, cat_col, cont_col, xlab, ylab):
         """Plot relation of categorical distribution with continuous."""
-        dist_type = self.util.config_val("EXPL", "dist_type", "hist")
+        dist_type = self.util.config_val("EXPL", "dist_type", "kde")
         cats, cat_str, es = su.get_effect_size(df, cat_col, cont_col)
         model_type = self.util.get_model_type()
         if dist_type == "hist" and model_type != "tree":

{nkululeko-0.89.1 → nkululeko-0.90.0}/nkululeko/reporting/reporter.py

@@ -402,7 +402,7 @@ class Reporter:
                 )
                 # print classifcation report in console
                 self.util.debug(
-                    f"\n {classification_report(self.truths, self.preds, target_names=labels)}"
+                    f"\n {classification_report(self.truths, self.preds, target_names=labels, digits=4)}"
                 )
             except ValueError as e:
                 self.util.debug(
@@ -422,16 +422,17 @@ class Reporter:
                 if len(np.unique(self.truths)) == 2:
                     fpr, tpr, _ = roc_curve(self.truths, self.preds)
                     auc_score = auc(fpr, tpr)
+                    plot_path = f"{fig_dir}{self.util.get_exp_name()}_{epoch}{self.filenameadd}_roc.{self.format}"
+                    plt.figure()
                     display = RocCurveDisplay(
                         fpr=fpr,
                         tpr=tpr,
                         roc_auc=auc_score,
                         estimator_name=f"{self.model_type} estimator",
                     )
-                    # save plot
-                    plot_path = f"{fig_dir}{self.util.get_exp_name()}_{epoch}{self.filenameadd}_roc.{self.format}"
                     display.plot(ax=None)
                     plt.savefig(plot_path)
+                    plt.close()
                     self.util.debug(f"Saved ROC curve to {plot_path}")
                     pauc_score = roc_auc_score(self.truths, self.preds, max_fpr=0.1)
                     auc_pauc = f"auc: {auc_score:.3f}, pauc: {pauc_score:.3f} from epoch: {epoch}"

{nkululeko-0.89.1 → nkululeko-0.90.0/nkululeko.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.89.1
+Version: 0.90.0
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -356,6 +356,16 @@ F. Burkhardt, Johannes Wagner, Hagen Wierstorf, Florian Eyben and Björn Schulle
 Changelog
 =========
 
+Version 0.90.0
+--------------
+* augmentation can now be done without target
+* random splicing params configurable
+* made kde default for plot continous/categorical plots
+
+Version 0.89.2
+--------------
+* fix shap value calculation
+  
 Version 0.89.1
 --------------
 * print and save result of feature importance