nkululeko 0.86.2__py3-none-any.whl → 0.86.4__py3-none-any.whl

nkululeko/constants.py

@@ -1,2 +1,2 @@
-VERSION="0.86.2"
+VERSION="0.86.4"
 SAMPLING_RATE = 16000

nkululeko/demo.py

@@ -20,16 +20,20 @@ Options:   \n
 import argparse
 import configparser
 import os
+import pandas as pd
 
 from nkululeko.constants import VERSION
 from nkululeko.experiment import Experiment
 import nkululeko.glob_conf as glob_conf
 from nkululeko.utils.util import Util
+from transformers import pipeline
 
 
 def main(src_dir):
-    parser = argparse.ArgumentParser(description="Call the nkululeko DEMO framework.")
-    parser.add_argument("--config", default="exp.ini", help="The base configuration")
+    parser = argparse.ArgumentParser(
+        description="Call the nkululeko DEMO framework.")
+    parser.add_argument("--config", default="exp.ini",
+                        help="The base configuration")
     parser.add_argument(
         "--file", help="A file that should be processed (16kHz mono wav)"
     )
@@ -79,6 +83,52 @@ def main(src_dir):
         f" {VERSION}"
     )
 
+    def print_pipe(files, outfile):
+        """
+        Prints the pipeline output for a list of files, and optionally writes the results to an output file.
+
+        Args:
+            files (list): A list of file paths to process through the pipeline.
+            outfile (str, optional): The path to an output file to write the pipeline results to.
+
+        Returns:
+            None
+        """
+        results = []
+        for file in files:
+            result = pipe(file, top_k=1)
+            if result[0]["score"] != result[0]["score"]:  # Check for NaN
+                print(f"ERROR: NaN value in pipeline output for file: {file}")
+            else:
+                results.append(f"{file}, {result[0]['label']}")
+        print("\n".join(results))
+
+        if outfile is not None:
+            with open(outfile, "w") as f:
+                f.write("\n".join(results))
+
+    if util.get_model_type() == "finetune":
+        model_path = os.path.join(
+            util.get_exp_dir(), "models", "run_0", "torch")
+        pipe = pipeline("audio-classification", model=model_path)
+        if args.file is not None:
+            print_pipe([args.file], args.outfile)
+        elif args.list is not None:
+            # read audio files from list
+            print(f"Reading files from {args.list}")
+            list_file = pd.read_csv(args.list, header="infer")
+            files = list_file.iloc[:, 0].tolist()
+            print_pipe(files, args.outfile)
+        elif args.folder is not None:
+            # read audio files from folder
+            from nkululeko.utils.files import find_files
+
+            files = find_files(args.folder, relative=True, ext=["wav", "mp3"])
+            print_pipe(files, args.outfile)
+        else:
+            print("ERROR: input mic currently is not supported for finetuning")
+        return
+
     # load the experiment
     expr.load(f"{util.get_save_name()}")
     if args.folder is not None:

nkululeko/experiment.py

@@ -115,7 +115,8 @@ class Experiment:
             self.util.debug(f"Target labels (from config): {labels}")
         else:
             self.labels = auto_labels
-            self.util.debug(f"Target labels (from database): {auto_labels}")
+        # print autolabel no matter it is specified or not
+        self.util.debug(f"Target labels (from database): {auto_labels}")
         glob_conf.set_labels(self.labels)
         self.util.debug(f"loaded databases {dbs}")
 

nkululeko/feat_extract/feats_import.py

@@ -35,6 +35,7 @@ class ImportSet(Featureset):
             if not os.path.isfile(feat_import_file):
                 self.util.error(f"no import file: {feat_import_file}")
             df = audformat.utils.read_csv(feat_import_file)
+            df = self.util.make_segmented_index(df)
             df = df[df.index.isin(self.data_df.index)]
             feat_df = pd.concat([feat_df, df])
         if feat_df.shape[0] == 0:

nkululeko/models/model_tuned.py

@@ -54,19 +54,23 @@ class TunedModel(BaseModel):
         self.learning_rate = float(
             self.util.config_val("MODEL", "learning_rate", "0.0001")
         )
-        self.max_duration = float(self.util.config_val("MODEL", "max_duration", "8.0"))
+        self.max_duration = float(
+            self.util.config_val("MODEL", "max_duration", "8.0"))
         self.df_train, self.df_test = df_train, df_test
         self.epoch_num = int(self.util.config_val("EXP", "epochs", 1))
+        self.util.debug(f"num of epochs: {self.epoch_num}")
         drop = self.util.config_val("MODEL", "drop", False)
         self.drop = 0.1
         if drop:
             self.drop = float(drop)
         self.util.debug(f"init: training with dropout: {self.drop}")
+        self.push = eval(self.util.config_val("MODEL", "push_to_hub", "False"))
         self._init_model()
 
     def _init_model(self):
         model_path = "facebook/wav2vec2-large-robust-ft-swbd-300h"
-        pretrained_model = self.util.config_val("MODEL", "pretrained_model", model_path)
+        pretrained_model = self.util.config_val(
+            "MODEL", "pretrained_model", model_path)
         self.num_layers = None
         self.sampling_rate = 16000
         self.max_duration_sec = self.max_duration
@@ -131,6 +135,10 @@ class TunedModel(BaseModel):
         tokenizer = transformers.Wav2Vec2CTCTokenizer("./vocab.json")
         tokenizer.save_pretrained(".")
 
+        # uoload tokenizer to hub if true
+        if self.push:
+            tokenizer.push_to_hub(self.util.get_name())
+
         feature_extractor = transformers.Wav2Vec2FeatureExtractor(
             feature_size=1,
             sampling_rate=16000,
@@ -260,7 +268,8 @@ class TunedModel(BaseModel):
                 else:
                     criterion = torch.nn.CrossEntropyLoss()
             else:
-                self.util.error(f"criterion {criterion} not supported for classifier")
+                self.util.error(
+                    f"criterion {criterion} not supported for classifier")
         else:
             self.criterion = self.util.config_val("MODEL", "loss", "ccc")
             if criterion == "1-ccc":
@@ -270,10 +279,11 @@ class TunedModel(BaseModel):
             elif criterion == "mae":
                 criterion = torch.nn.L1Loss()
             else:
-                self.util.error(f"criterion {criterion} not supported for regressor")
+                self.util.error(
+                    f"criterion {criterion} not supported for regressor")
 
         # set push_to_hub value, default false
-        push = eval(self.util.config_val("MODEL", "push_to_hub", "False"))
+        # push = eval(self.util.config_val("MODEL", "push_to_hub", "False"))
 
         class Trainer(transformers.Trainer):
             def compute_loss(
@@ -309,7 +319,8 @@ class TunedModel(BaseModel):
         elif metrics_for_best_model == "MAE":
             greater_is_better = False
         else:
-            self.util.error(f"unknown metric/measure: {metrics_for_best_model}")
+            self.util.error(
+                f"unknown metric/measure: {metrics_for_best_model}")
 
         training_args = transformers.TrainingArguments(
             output_dir=model_root,
@@ -319,7 +330,8 @@ class TunedModel(BaseModel):
             gradient_accumulation_steps=self.accumulation_steps,
             evaluation_strategy="steps",
             num_train_epochs=self.epoch_num,
-            fp16=self.device == "cuda",
+            fp16=self.device != "cpu",
+            use_cpu=self.device == "cpu",
             save_steps=num_steps,
             eval_steps=num_steps,
             logging_steps=num_steps,
@@ -331,8 +343,9 @@ class TunedModel(BaseModel):
             load_best_model_at_end=True,
             remove_unused_columns=False,
             report_to="none",
-            push_to_hub=push,
+            push_to_hub=self.push,
             hub_model_id=f"{self.util.get_name()}",
+            overwrite_output_dir=True,
         )
 
         trainer = Trainer(
@@ -440,7 +453,7 @@ class TunedModel(BaseModel):
             self.clf = pickle.load(handle)
 
 
-@dataclasses.dataclass
+@ dataclasses.dataclass
 class ModelOutput(transformers.file_utils.ModelOutput):
 
     logits: torch.FloatTensor = None
@@ -448,7 +461,7 @@ class ModelOutput(transformers.file_utils.ModelOutput):
     cnn_features: torch.FloatTensor = None
 
 
-@dataclasses.dataclass
+@ dataclasses.dataclass
 class ModelOutputReg(transformers.file_utils.ModelOutput):
 
     logits: torch.FloatTensor
@@ -517,7 +530,10 @@ class Model(Wav2Vec2PreTrainedModel):
             )
             outputs = torch.sum(hidden_states, dim=1)
             attention_sum = torch.sum(attention_mask, dim=1)
-            outputs = outputs / torch.reshape(attention_sum, (-1, 1))
+
+            epsilon = 1e-6  # to avoid division by zero and numerical instability
+            outputs = outputs / (torch.reshape(attention_sum, (-1, 1)) +
+                                 epsilon)
 
         return outputs
 

nkululeko/reporting/reporter.py

@@ -315,7 +315,7 @@ class Reporter:
         plt.savefig(plot_path)
         self.util.debug(f"plotted epoch progression to {plot_path}")
         plt.close(fig)
-        fig.clear()
+        # fig.clear()
 
     def plot_epoch_progression(self, reports, out_name):
         fig_dir = self.util.get_path("fig_dir")

nkululeko/utils/util.py

@@ -35,6 +35,7 @@ class Util:
         if has_config:
             try:
                 import nkululeko.glob_conf as glob_conf
+
                 self.config = glob_conf.config
                 self.got_data_roots = self.config_val(
                     "DATA", "root_folders", False)
@@ -160,6 +161,8 @@ class Util:
         if len(df) == 0:
             return df
         if not isinstance(df.index, pd.MultiIndex):
+            self.debug(
+                "converting to segmented index, this might take a while...")
             df.index = audformat.utils.to_segmented_index(
                 df.index, allow_nat=False)
         return df
@@ -196,6 +199,9 @@ class Util:
             return_string = return_string + "_" + mt
         return return_string.replace("__", "_")
 
+    def get_model_type(self):
+        return self.config["MODEL"]["type"]
+
     def get_model_description(self):
         mt = ""
         mt = f'{self.config["MODEL"]["type"]}'

{nkululeko-0.86.2.dist-info → nkululeko-0.86.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: nkululeko
-Version: 0.86.2
+Version: 0.86.4
 Summary: Machine learning audio prediction experiments based on templates
 Home-page: https://github.com/felixbur/nkululeko
 Author: Felix Burkhardt
@@ -237,6 +237,7 @@ There's my [blog](http://blog.syntheticspeech.de/?s=nkululeko) with tutorials:
 * [Run multiple experiments in one go](http://blog.syntheticspeech.de/2022/03/28/how-to-run-multiple-experiments-in-one-go-with-nkululeko/)
 * [Compare several MLP layer layouts with each other](http://blog.syntheticspeech.de/2022/04/11/how-to-compare-several-mlp-layer-layouts-with-each-other/)
 * [Import features from outside the software](http://blog.syntheticspeech.de/2022/10/18/how-to-import-features-from-outside-the-nkululeko-software/)
+* [Export acoustic features](http://blog.syntheticspeech.de/2024/05/30/nkululeko-export-acoustic-features/)
 * [Explore feature importance](http://blog.syntheticspeech.de/2023/02/20/nkululeko-show-feature-importance/)
 * [Plot distributions for feature values](http://blog.syntheticspeech.de/2023/02/16/nkululeko-how-to-plot-distributions-of-feature-values/)
 * [Show feature importance](http://blog.syntheticspeech.de/2023/02/20/nkululeko-show-feature-importance/)
@@ -334,6 +335,15 @@ F. Burkhardt, Johannes Wagner, Hagen Wierstorf, Florian Eyben and Björn Schulle
 Changelog
 =========
 
+Version 0.86.4
+--------------
+* add finetuning to the demo module
+
+Version 0.86.3
+--------------
+* bugfixed: nan in finetuned model and double saving
+* import features now get multiindex automatically
+
 Version 0.86.2
 --------------
 * plots epoch progression for finetuned models now

{nkululeko-0.86.2.dist-info → nkululeko-0.86.4.dist-info}/RECORD

@@ -2,11 +2,11 @@ nkululeko/__init__.py,sha256=62f8HiEzJ8rG2QlTFJXUCMpvuH3fKI33DoJSj33mscc,63
 nkululeko/aug_train.py,sha256=YhuZnS_WVWnun9G-M6g5n6rbRxoVREz6Zh7k6qprFNQ,3194
 nkululeko/augment.py,sha256=4MG0apTAG5RgkuJrYEjGgDdbodZWi_HweSPNI1JJ5QA,3051
 nkululeko/cacheddataset.py,sha256=lIJ6hUo5LoxSrzXtWV8mzwO7wRtUETWnOQ4ws2XfL1E,969
-nkululeko/constants.py,sha256=E9mXpvAI5IDamRRXgBlBH8XGTw1xEjEBzNibjhFPEFc,39
-nkululeko/demo.py,sha256=8bl15Kitoesnz8oa8yrs52T6YCSOhWbbq9PnZ8Hj6D0,3232
+nkululeko/constants.py,sha256=QtEoU6iCjnUpcJT-FOh4bU4miJ_D0z26OpSub4oEY1c,39
+nkululeko/demo.py,sha256=WSKr-W5uJ9DQfemK923g7Hd5V3kgAn03Er0JX1Pa45I,5142
 nkululeko/demo_feats.py,sha256=sAeGFojhEj9WEDFtG3SzPBmyYJWLF2rkbpp65m8Ujo4,2025
 nkululeko/demo_predictor.py,sha256=es56xbT8ifkS_vnrlb5NTZT54gNmeUtNlA4zVA_gnN8,4757
-nkululeko/experiment.py,sha256=24FmvF9_zNXE86fO6gzss1M-BjceOCiV6nyJAs0SM_Y,30986
+nkululeko/experiment.py,sha256=huhHLQfnzxRJlQi2SY61XMWbC8xEWpe31yq9spBUk-4,31041
 nkululeko/explore.py,sha256=lDzRoW_Taa5u4BBABZLD89BcQWnYlrftJR4jgt1yyj0,2609
 nkululeko/export.py,sha256=mHeEAAmtZuxdyebLlbSzPrHSi9OMgJHbk35d3DTxRBc,4632
 nkululeko/feature_extractor.py,sha256=8mssYKmo4LclVI-hiLmJEDZ0ZPyDavFG2YwtXcrGzwM,3976
@@ -55,7 +55,7 @@ nkululeko/feat_extract/feats_auddim.py,sha256=VlzKKXTXa5kjLgQBWyEFy-daIyU1SkOwCC
 nkululeko/feat_extract/feats_audmodel.py,sha256=VjBNgAoxsHJhwr6Kwt9CxX6SaCM4RK_OV-GU2W5-bhU,3187
 nkululeko/feat_extract/feats_clap.py,sha256=nR6eEIRdsMHcfmD1bNtt5WfDvkxKjvEbukSSrXHm-HU,3489
 nkululeko/feat_extract/feats_hubert.py,sha256=cLoUzSLjSYBkQnftjacSL7ES3O7Ysh_KrPYvZtLX_TU,5196
-nkululeko/feat_extract/feats_import.py,sha256=rj1p8lz19tCAC8hLzzZAwZ0M6gzwH3BzfabFUgal0yw,1622
+nkululeko/feat_extract/feats_import.py,sha256=WiU5lCkJsmFNTDyPV0qIh8mJssa6bpgP7AYw_ClKfWM,1674
 nkululeko/feat_extract/feats_mld.py,sha256=Vvu7GZOkn7Vda8eIOXqHjg78zegkFe3vTUaCXyVM0eA,2021
 nkululeko/feat_extract/feats_mos.py,sha256=KXNt7QYEfxkvr6UyVhig2aWQBaIvovlrR4gPuP03gmo,4174
 nkululeko/feat_extract/feats_opensmile.py,sha256=g6ZsAxjjGGvGfrr5fngWC-NJ8E7CP1kYZwrlodZJzzU,4028
@@ -88,7 +88,7 @@ nkululeko/models/model_svm.py,sha256=rsME3KvKvNG7bdE5lbvYUu85WZhaASZxxmdNDIVJRZ4
 nkululeko/models/model_svr.py,sha256=_YZeksqB3eBENGlg3g9RwYFlk9rQQ-XCeNBKLlGGVoE,725
 nkululeko/models/model_tree.py,sha256=rf16faUm4o2LJgkoYpeY998b8DQIvXZ73_m1IS3TnnE,417
 nkululeko/models/model_tree_reg.py,sha256=IgQcPTE-304HQLYSKPF8Z4ot_Ur9dH01fZjS0nXke_M,428
-nkululeko/models/model_tuned.py,sha256=xOoY5TROzzTVu3sDtlmEle3V1MAgpf8S3WxO9o4MzV4,20777
+nkululeko/models/model_tuned.py,sha256=vmNBkqvEH-4nnhY1REXDA9kA4vpZJzeRmGJFq7E3bLM,21340
 nkululeko/models/model_xgb.py,sha256=Thgx5ESdIok4v72mKh4plxpo4smGcKALWNCJTDScY0M,447
 nkululeko/models/model_xgr.py,sha256=aGBtNGLWjOE_2rICGYGFxmT8DtnHYsIl1lIpMtghHsY,418
 nkululeko/reporting/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -96,7 +96,7 @@ nkululeko/reporting/defines.py,sha256=IsY1YgKRMaABpylVKjBJgJ5bNCEbGCVA_E6pivraqS
 nkululeko/reporting/latex_writer.py,sha256=qiCRSmB4KOD_za4oHu5x-PhwjZohzfo8wecMOwlXZwc,1886
 nkululeko/reporting/report.py,sha256=W0rcigDdjBvxZQ3pZja_gvToILYvaZ1BFtnN2qFRfYI,1060
 nkululeko/reporting/report_item.py,sha256=siWeGNgo4bAE46YBMNcsdf3jTMTy76BO9Fi6DTvDig4,533
-nkululeko/reporting/reporter.py,sha256=NugmGmS3iwuBJ59jqyuTCKPRpiPLGhnz12z_nlVh69Y,13445
+nkululeko/reporting/reporter.py,sha256=4dXRwJ-CZ49NlF_kv9hfDjZT3bbWNNMyNpKEBLKs3Ew,13447
 nkululeko/reporting/result.py,sha256=nSN5or-Py2GPRWHkWpGRh7UCi1W0er7WLEHz8fYLk-A,742
 nkululeko/segmenting/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nkululeko/segmenting/seg_inaspeechsegmenter.py,sha256=pmLHuXsaqvcdYxB4PSW9l1mbQWZZBJFhi_CGabqydas,1947
@@ -104,9 +104,9 @@ nkululeko/segmenting/seg_silero.py,sha256=lLytS38KzARS17omwv8VBw-zz60RVSXGSvZ5Ev
 nkululeko/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nkululeko/utils/files.py,sha256=UiGAtZRWYjHSvlmPaTMtzyNNGE6qaLaxQkybctS7iRM,4021
 nkululeko/utils/stats.py,sha256=1yUq0FTOyqkU8TwUocJRYdJaqMU5SlOBBRUun9STo2M,2829
-nkululeko/utils/util.py,sha256=mK1MgO14NinrPhavJw72eR_2WN_kBKjVKiEJnzvdO1Q,13946
-nkululeko-0.86.2.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
-nkululeko-0.86.2.dist-info/METADATA,sha256=DmmpMrftBptpWqx7h9US7_4mvMIQbZ5ugzv_4kyBjkM,37170
-nkululeko-0.86.2.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
-nkululeko-0.86.2.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
-nkululeko-0.86.2.dist-info/RECORD,,
+nkululeko/utils/util.py,sha256=ILpfNuaeq-hy1bUkRhVrzO2wG9z9Upaozs9EBoIaMG0,14123
+nkululeko-0.86.4.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
+nkululeko-0.86.4.dist-info/METADATA,sha256=D1y8wrwDr0gLVdafV4E_GcER5yrt3IaKUdqJ8huMCwA,37480
+nkululeko-0.86.4.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
+nkululeko-0.86.4.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
+nkululeko-0.86.4.dist-info/RECORD,,