nkululeko 0.90.1__py3-none-any.whl → 0.90.2__py3-none-any.whl

nkululeko/constants.py

@@ -1,2 +1,2 @@
-VERSION = "0.90.1"
+VERSION = "0.90.2"
 SAMPLING_RATE = 16000

nkululeko/models/model_tuned.py

@@ -30,10 +30,16 @@ class TunedModel(BaseModel):
         """Constructor taking the configuration and all dataframes."""
         super().__init__(df_train, df_test, feats_train, feats_test)
         super().set_model_type("finetuned")
+        self.df_test, self.df_train, self.feats_test, self.feats_train = (
+            df_test,
+            df_train,
+            feats_test,
+            feats_train,
+        )
         self.name = "finetuned_wav2vec2"
         self.target = glob_conf.config["DATA"]["target"]
-        labels = glob_conf.labels
-        self.class_num = len(labels)
+        self.labels = glob_conf.labels
+        self.class_num = len(self.labels)
         device = self.util.config_val("MODEL", "device", False)
         if not device:
             self.device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -304,7 +310,7 @@ class TunedModel(BaseModel):
             else:
                 self.util.error(f"criterion {criterion} not supported for classifier")
         else:
-            self.criterion = self.util.config_val("MODEL", "loss", "ccc")
+            criterion = self.util.config_val("MODEL", "loss", "1-ccc")
             if criterion == "1-ccc":
                 criterion = ConcordanceCorCoeff()
             elif criterion == "mse":
@@ -402,7 +408,7 @@ class TunedModel(BaseModel):
         self.load(self.run, self.epoch)
 
     def get_predictions(self):
-        results = []
+        results = [[]].pop(0)
         for (file, start, end), _ in audeer.progress_bar(
             self.df_test.iterrows(),
             total=len(self.df_test),
@@ -415,18 +421,37 @@ class TunedModel(BaseModel):
                     file, duration=end - start, offset=start, always_2d=True
                 )
             assert sr == self.sampling_rate
-            predictions = self.model.predict(signal)
-            results.append(predictions.argmax())
-        return results
+            prediction = self.model.predict(signal)
+            results.append(prediction)
+            # results.append(predictions.argmax())
+        predictions = np.asarray(results)
+        if self.util.exp_is_classification():
+            # make a dataframe for the class probabilities
+            proba_d = {}
+            for c in range(self.class_num):
+                proba_d[c] = []
+            # get the class probabilities
+            # predictions = self.clf.predict_proba(self.feats_test.to_numpy())
+            # pred = self.clf.predict(features)
+            for i in range(self.class_num):
+                proba_d[i] = list(predictions.T[i])
+            probas = pd.DataFrame(proba_d)
+            probas = probas.set_index(self.df_test.index)
+            predictions = probas.idxmax(axis=1).values
+        else:
+            predictions = predictions.flatten()
+            probas = None
+        return predictions, probas
 
     def predict(self):
         """Predict the whole eval feature set"""
-        predictions = self.get_predictions()
+        predictions, probas = self.get_predictions()
         report = Reporter(
             self.df_test[self.target].to_numpy().astype(float),
             predictions,
             self.run,
             self.epoch_num,
+            probas=probas,
         )
         self._plot_epoch_progression(report)
         return report
@@ -438,6 +463,7 @@ class TunedModel(BaseModel):
         )
         with open(log_file, "r") as file:
             data = file.read()
+        data = data.strip().replace("nan", "0")
         list = ast.literal_eval(data)
         epochs, vals, loss = [], [], []
         for index, tp in enumerate(list):

nkululeko/utils/util.py

@@ -155,10 +155,10 @@ class Util:
         return f"{store}/{self.get_exp_name()}.pkl"
 
     def get_pred_name(self):
-        store = self.get_path("store")
+        results_dir = self.get_path("res_dir")
         target = self.get_target_name()
         pred_name = self.get_model_description()
-        return f"{store}/pred_{target}_{pred_name}.csv"
+        return f"{results_dir}/pred_{target}_{pred_name}.csv"
 
     def is_categorical(self, pd_series):
         """Check if a dataframe column is categorical."""

{nkululeko-0.90.1.dist-info → nkululeko-0.90.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.90.1
+Version: 0.90.2
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -356,6 +356,11 @@ F. Burkhardt, Johannes Wagner, Hagen Wierstorf, Florian Eyben and Björn Schulle
 Changelog
 =========
 
+Version 0.90.2
+--------------
+* added probability output to finetuning classification models
+* switched path to prob. output from "store" to "results"
+
 Version 0.90.1
 --------------
 * Add balancing for finetune and update data README

{nkululeko-0.90.1.dist-info → nkululeko-0.90.2.dist-info}/RECORD

@@ -2,7 +2,7 @@ nkululeko/__init__.py,sha256=62f8HiEzJ8rG2QlTFJXUCMpvuH3fKI33DoJSj33mscc,63
 nkululeko/aug_train.py,sha256=FoMbBrfyOZd4QAw7oIHl3X6-UpsqAKWVDIolCA7qOWs,3196
 nkululeko/augment.py,sha256=sIXRg19Uz8dWKgQv2LBGH7jbd2pgcUTh0PIQ_62B0kA,3135
 nkululeko/cacheddataset.py,sha256=XFpWZmbJRg0pvhnIgYf0TkclxllD-Fctu-Ol0PF_00c,969
-nkululeko/constants.py,sha256=TmPPFi_-OUMYF2mfBNMLxBQl0vwneI1opUPN0vK2XPY,41
+nkululeko/constants.py,sha256=RbyLuq3HuWP1QWBrcWXo-YcwlYf2qDk6H1ihR4_KqbY,41
 nkululeko/demo-ft.py,sha256=iD9Pzp9QjyAv31q1cDZ75vPez7Ve8A4Cfukv5yfZdrQ,770
 nkululeko/demo.py,sha256=bLuHkeEl5rOfm7ecGHCcWATiPK7-njNbtrGljxzNzFs,5088
 nkululeko/demo_feats.py,sha256=BvZjeNFTlERIRlq34OHM4Z96jdDQAhB01BGQAUcX9dM,2026
@@ -95,7 +95,7 @@ nkululeko/models/model_svm.py,sha256=zP8ykLhCZTYvwSqw06XHuzq9qMBtsiYpxjUpWDAnMyA
 nkululeko/models/model_svr.py,sha256=FEwYRdgqwgGhZdkpRnT7Ef12lklWi6GZL28PyV99xWs,726
 nkululeko/models/model_tree.py,sha256=6L3PD3aIiiQz1RPWS6z3Edx4f0gnR7AOfBKOJzf0BNU,433
 nkululeko/models/model_tree_reg.py,sha256=IMaQpNImoRqP8Biw1CsJevxpV_PVpKblsKtYlMW5d_U,429
-nkululeko/models/model_tuned.py,sha256=k6c8dPKy2BeFMKABrNTMSwQuiKa9VrZ7oeJdfNYoYAo,22678
+nkululeko/models/model_tuned.py,sha256=VuRyNqw3XTpQ2eHsWOJN8X-V98AN8Wqiq7UgwT5BQRU,23763
 nkululeko/models/model_xgb.py,sha256=ytBaSHZH8r7VvRYdmrBrQnzRM6V4HyCJ8O-v20J8G_g,448
 nkululeko/models/model_xgr.py,sha256=H01FJCRgmX2unvambMs5TTCS9sI6VDB9ip9G6rVGt2c,419
 nkululeko/reporting/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -111,9 +111,9 @@ nkululeko/segmenting/seg_silero.py,sha256=CnhjKGTW5OXf-bmw4YsSJeN2yUwkY5m3xnulM_
 nkululeko/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nkululeko/utils/files.py,sha256=UiGAtZRWYjHSvlmPaTMtzyNNGE6qaLaxQkybctS7iRM,4021
 nkululeko/utils/stats.py,sha256=vCRzhCR0Gx5SiJyAGbj1TIto8ocGz58CM5Pr3LltagA,2948
-nkululeko/utils/util.py,sha256=a9fs5swVkv_k0CfJRwDhEx1ChZv7rs7K4oQDYspiQWY,16709
-nkululeko-0.90.1.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
-nkululeko-0.90.1.dist-info/METADATA,sha256=unqq8xrL0bfP178Q3fKBaGyry4SJvHxPGJCR3figOpQ,40961
-nkululeko-0.90.1.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
-nkululeko-0.90.1.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
-nkululeko-0.90.1.dist-info/RECORD,,
+nkululeko/utils/util.py,sha256=XFZdhCc_LM4EmoZ5tKKaBCQLXclcNmvHwhfT_CXB98c,16723
+nkululeko-0.90.2.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
+nkululeko-0.90.2.dist-info/METADATA,sha256=rJnGf71UEIyv0OBiNxrfu0l1e6o83v8q_UlIlmhtE_0,41113
+nkululeko-0.90.2.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
+nkululeko-0.90.2.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
+nkululeko-0.90.2.dist-info/RECORD,,