nkululeko 0.82.4__py3-none-any.whl → 0.83.1__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
nkululeko/constants.py CHANGED
@@ -1,2 +1,2 @@
1
- VERSION="0.82.4"
1
+ VERSION="0.83.1"
2
2
  SAMPLING_RATE = 16000
nkululeko/experiment.py CHANGED
@@ -23,7 +23,7 @@ from nkululeko.plots import Plots
23
23
  from nkululeko.reporting.report import Report
24
24
  from nkululeko.runmanager import Runmanager
25
25
  from nkululeko.scaler import Scaler
26
- from nkululeko.test_predictor import Test_predictor
26
+ from nkululeko.test_predictor import TestPredictor
27
27
  from nkululeko.utils.util import Util
28
28
 
29
29
 
@@ -672,10 +672,11 @@ class Experiment:
672
672
  def predict_test_and_save(self, result_name):
673
673
  model = self.runmgr.get_best_model()
674
674
  model.set_testdata(self.df_test, self.feats_test)
675
- test_predictor = Test_predictor(
675
+ test_predictor = TestPredictor(
676
676
  model, self.df_test, self.label_encoder, result_name
677
677
  )
678
- test_predictor.predict_and_store()
678
+ result = test_predictor.predict_and_store()
679
+ return result
679
680
 
680
681
  def load(self, filename):
681
682
  f = open(filename, "rb")
nkululeko/modelrunner.py CHANGED
@@ -2,18 +2,16 @@
2
2
 
3
3
  import pandas as pd
4
4
 
5
- from nkululeko.utils.util import Util
6
5
  from nkululeko import glob_conf
7
- import nkululeko.glob_conf as glob_conf
6
+ from nkululeko.utils.util import Util
8
7
 
9
8
 
10
9
  class Modelrunner:
11
- """
12
- Class to model one run
13
- """
10
+ """Class to model one run."""
14
11
 
15
12
  def __init__(self, df_train, df_test, feats_train, feats_test, run):
16
- """Constructor setting up the dataframes
13
+ """Constructor setting up the dataframes.
14
+
17
15
  Args:
18
16
  df_train: train dataframe
19
17
  df_test: test dataframe
nkululeko/nkuluflag.py CHANGED
@@ -2,13 +2,16 @@ import argparse
2
2
  import configparser
3
3
  import os
4
4
  import os.path
5
+ import sys
5
6
 
6
7
  from nkululeko.nkululeko import doit as nkulu
8
+ from nkululeko.test import do_it as test_mod
7
9
 
8
10
 
9
- def do_it(src_dir):
11
+ def doit(cla):
10
12
  parser = argparse.ArgumentParser(description="Call the nkululeko framework.")
11
13
  parser.add_argument("--config", help="The base configuration")
14
+ parser.add_argument("--mod", default="nkulu", help="Which nkululeko module to call")
12
15
  parser.add_argument("--data", help="The databases", nargs="*", action="append")
13
16
  parser.add_argument(
14
17
  "--label", nargs="*", help="The labels for the target", action="append"
@@ -25,20 +28,23 @@ def do_it(src_dir):
25
28
  parser.add_argument("--model", default="xgb", help="The model type")
26
29
  parser.add_argument("--feat", default="['os']", help="The feature type")
27
30
  parser.add_argument("--set", help="The opensmile set")
28
- parser.add_argument("--with_os", help="To add os features")
29
31
  parser.add_argument("--target", help="The target designation")
30
32
  parser.add_argument("--epochs", help="The number of epochs")
31
33
  parser.add_argument("--runs", help="The number of runs")
32
34
  parser.add_argument("--learning_rate", help="The learning rate")
33
35
  parser.add_argument("--drop", help="The dropout rate [0:1]")
34
36
 
35
- args = parser.parse_args()
37
+ args = parser.parse_args(cla)
36
38
 
37
39
  if args.config is not None:
38
40
  config_file = args.config
39
41
  else:
40
42
  print("ERROR: need config file")
41
43
  quit(-1)
44
+
45
+ if args.mod is not None:
46
+ nkulu_mod = args.mod
47
+
42
48
  # test if config is there
43
49
  if not os.path.isfile(config_file):
44
50
  print(f"ERROR: no such file {config_file}")
@@ -86,10 +92,17 @@ def do_it(src_dir):
86
92
  with open(tmp_config, "w") as tmp_file:
87
93
  config.write(tmp_file)
88
94
 
89
- result, last_epoch = nkulu(tmp_config)
95
+ result, last_epoch = 0, 0
96
+ if nkulu_mod == "nkulu":
97
+ result, last_epoch = nkulu(tmp_config)
98
+ elif nkulu_mod == "test":
99
+ result, last_epoch = test_mod(tmp_config, "test_results.csv")
100
+ else:
101
+ print(f"ERROR: unknown module: {nkulu_mod}, should be [nkulu | test]")
90
102
  return result, last_epoch
91
103
 
92
104
 
93
105
  if __name__ == "__main__":
94
- cwd = os.path.dirname(os.path.abspath(__file__))
95
- do_it(cwd) # sys.argv[1])
106
+ cla = sys.argv
107
+ cla.pop(0)
108
+ doit(cla) # sys.argv[1])
@@ -55,6 +55,7 @@ class Reporter:
55
55
  self.run = run
56
56
  self.epoch = epoch
57
57
  self.__set_measure()
58
+ self.filenameadd = ""
58
59
  self.cont_to_cat = False
59
60
  if len(self.truths) > 0 and len(self.preds) > 0:
60
61
  if self.util.exp_is_classification():
@@ -206,7 +207,7 @@ class Reporter:
206
207
  f"Confusion Matrix, UAR: {uar_str} "
207
208
  + f"(+-{up_str}/{low_str}) {reg_res}"
208
209
  )
209
- img_path = f"{fig_dir}{plot_name}.{self.format}"
210
+ img_path = f"{fig_dir}{plot_name}{self.filenameadd}.{self.format}"
210
211
  plt.savefig(img_path)
211
212
  fig.clear()
212
213
  plt.close(fig)
@@ -228,14 +229,17 @@ class Reporter:
228
229
  )
229
230
  # print(rpt)
230
231
  self.util.debug(rpt)
231
- file_name = f"{res_dir}{self.util.get_exp_name()}_conf.txt"
232
+ file_name = f"{res_dir}{self.util.get_exp_name()}{self.filenameadd}_conf.txt"
232
233
  with open(file_name, "w") as text_file:
233
234
  text_file.write(rpt)
234
235
 
236
+ def set_filename_add(self, my_string):
237
+ self.filenameadd = f"_{my_string}"
238
+
235
239
  def print_results(self, epoch):
236
240
  """Print all evaluation values to text file."""
237
241
  res_dir = self.util.get_path("res_dir")
238
- file_name = f"{res_dir}{self.util.get_exp_name()}_{epoch}.txt"
242
+ file_name = f"{res_dir}{self.util.get_exp_name()}_{epoch}{self.filenameadd}.txt"
239
243
  if self.util.exp_is_classification():
240
244
  labels = glob_conf.labels
241
245
  try:
nkululeko/test.py CHANGED
@@ -10,20 +10,7 @@ from nkululeko.experiment import Experiment
10
10
  from nkululeko.utils.util import Util
11
11
 
12
12
 
13
- def main(src_dir):
14
- parser = argparse.ArgumentParser(
15
- description="Call the nkululeko TEST framework.")
16
- parser.add_argument("--config", default="exp.ini",
17
- help="The base configuration")
18
- parser.add_argument(
19
- "--outfile",
20
- default="my_results.csv",
21
- help="File name to store the predictions",
22
- )
23
-
24
- args = parser.parse_args()
25
-
26
- config_file = args.config
13
+ def do_it(config_file, outfile):
27
14
 
28
15
  # test if the configuration file exists
29
16
  if not os.path.isfile(config_file):
@@ -48,10 +35,28 @@ def main(src_dir):
48
35
  expr.load(f"{util.get_save_name()}")
49
36
  expr.fill_tests()
50
37
  expr.extract_test_feats()
51
- expr.predict_test_and_save(args.outfile)
38
+ result = expr.predict_test_and_save(outfile)
52
39
 
53
40
  print("DONE")
54
41
 
42
+ return result, 0
43
+
44
+
45
+ def main(src_dir):
46
+ parser = argparse.ArgumentParser(description="Call the nkululeko TEST framework.")
47
+ parser.add_argument("--config", default="exp.ini", help="The base configuration")
48
+ parser.add_argument(
49
+ "--outfile",
50
+ default="my_results.csv",
51
+ help="File name to store the predictions",
52
+ )
53
+ args = parser.parse_args()
54
+ if args.config is not None:
55
+ config_file = args.config
56
+ else:
57
+ config_file = f"{src_dir}/exp.ini"
58
+ do_it(config_file, args.outfile)
59
+
55
60
 
56
61
  if __name__ == "__main__":
57
62
  cwd = os.path.dirname(os.path.abspath(__file__))
@@ -1,21 +1,25 @@
1
- """ test_predictor.py
1
+ """test_predictor.py.
2
+
2
3
  Predict targets from a model and save as csv file.
3
4
 
4
5
  """
5
6
 
6
- import nkululeko.glob_conf as glob_conf
7
- from nkululeko.utils.util import Util
7
+ import ast
8
+
9
+ import numpy as np
8
10
  import pandas as pd
11
+ from sklearn.preprocessing import LabelEncoder
12
+
9
13
  from nkululeko.data.dataset import Dataset
10
14
  from nkululeko.feature_extractor import FeatureExtractor
15
+ import nkululeko.glob_conf as glob_conf
11
16
  from nkululeko.scaler import Scaler
12
- import numpy as np
13
- from sklearn.preprocessing import LabelEncoder
17
+ from nkululeko.utils.util import Util
14
18
 
15
19
 
16
- class Test_predictor:
20
+ class TestPredictor:
17
21
  def __init__(self, model, orig_df, labenc, name):
18
- """Constructor setting up name and configuration"""
22
+ """Constructor setting up name and configuration."""
19
23
  self.model = model
20
24
  self.orig_df = orig_df
21
25
  self.label_encoder = labenc
@@ -25,6 +29,7 @@ class Test_predictor:
25
29
 
26
30
  def predict_and_store(self):
27
31
  label_data = self.util.config_val("DATA", "label_data", False)
32
+ result = 0
28
33
  if label_data:
29
34
  data = Dataset(label_data)
30
35
  data.load()
@@ -49,7 +54,15 @@ class Test_predictor:
49
54
  df[self.target] = labelenc.inverse_transform(predictions.tolist())
50
55
  df.to_csv(self.name)
51
56
  else:
57
+ test_dbs = ast.literal_eval(glob_conf.config["DATA"]["tests"])
58
+ test_dbs_string = "_".join(test_dbs)
52
59
  predictions = self.model.get_predictions()
60
+ report = self.model.predict()
61
+ result = report.result.get_result()
62
+ report.set_filename_add(f"test-{test_dbs_string}")
63
+ self.util.print_best_results([report])
64
+ report.plot_confmatrix(self.util.get_plot_name(), 0)
65
+ report.print_results(0)
53
66
  # print(predictions)
54
67
  # df = pd.DataFrame(index=self.orig_df.index)
55
68
  # df["speaker"] = self.orig_df["speaker"]
@@ -63,3 +76,4 @@ class Test_predictor:
63
76
  df = df.rename(columns={"class_label": target})
64
77
  df.to_csv(self.name)
65
78
  self.util.debug(f"results stored in {self.name}")
79
+ return result
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.1
2
2
  Name: nkululeko
3
- Version: 0.82.4
3
+ Version: 0.83.1
4
4
  Summary: Machine learning audio prediction experiments based on templates
5
5
  Home-page: https://github.com/felixbur/nkululeko
6
6
  Author: Felix Burkhardt
@@ -333,6 +333,14 @@ F. Burkhardt, Johannes Wagner, Hagen Wierstorf, Florian Eyben and Björn Schulle
333
333
  Changelog
334
334
  =========
335
335
 
336
+ Version 0.83.1
337
+ --------------
338
+ * add test module to nkuluflag
339
+
340
+ Version 0.83.0
341
+ --------------
342
+ * test module now prints out reports
343
+
336
344
  Version 0.82.4
337
345
  --------------
338
346
  * fixed bug in wavlm
@@ -2,31 +2,30 @@ nkululeko/__init__.py,sha256=62f8HiEzJ8rG2QlTFJXUCMpvuH3fKI33DoJSj33mscc,63
2
2
  nkululeko/aug_train.py,sha256=YhuZnS_WVWnun9G-M6g5n6rbRxoVREz6Zh7k6qprFNQ,3194
3
3
  nkululeko/augment.py,sha256=4MG0apTAG5RgkuJrYEjGgDdbodZWi_HweSPNI1JJ5QA,3051
4
4
  nkululeko/cacheddataset.py,sha256=lIJ6hUo5LoxSrzXtWV8mzwO7wRtUETWnOQ4ws2XfL1E,969
5
- nkululeko/constants.py,sha256=kp-581CniawYq15MP-eOYf2iEPJCoCdhJ8W3kRUgPhs,39
5
+ nkululeko/constants.py,sha256=i6-Vtyje9xE8w8o3lG27IiJczQFyrNbsxiXs7b4-q28,39
6
6
  nkululeko/demo.py,sha256=55kNFA2helMhOxD4yZuKg1JWDtlUUpxm-6uAnroIydI,3264
7
7
  nkululeko/demo_feats.py,sha256=sAeGFojhEj9WEDFtG3SzPBmyYJWLF2rkbpp65m8Ujo4,2025
8
8
  nkululeko/demo_predictor.py,sha256=-ggSHc3DXxRzjzcGB4qFBOMvKsfUdTkkde50BDrS9dA,4755
9
- nkululeko/experiment.py,sha256=CSEvQxK2_tzJyND5sUHQSc6MkRp1g6EVam8JX8txqps,29576
9
+ nkululeko/experiment.py,sha256=aueWoKJCQx8wU9daosh6n7ZDGhT2cfo_9Av5HIfN1_w,29605
10
10
  nkululeko/explore.py,sha256=2wdoGRqldvsN1zCiWk0quSDgHHHUoF2UZOWQ1r-2OLM,2310
11
11
  nkululeko/export.py,sha256=mHeEAAmtZuxdyebLlbSzPrHSi9OMgJHbk35d3DTxRBc,4632
12
12
  nkululeko/feature_extractor.py,sha256=8mssYKmo4LclVI-hiLmJEDZ0ZPyDavFG2YwtXcrGzwM,3976
13
13
  nkululeko/file_checker.py,sha256=LoLnL8aHpW-axMQ46qbqrManTs5otG9ShpEZuz9iRSk,3474
14
14
  nkululeko/filter_data.py,sha256=w-X2mhKdYr5DxDIz50E5yzO6Jmzk4jjDBoXsgOOVtcA,7222
15
15
  nkululeko/glob_conf.py,sha256=iHiVSxDYgmYwdx6z0HuGUMSWrfZfufPHxHb60q2dLRY,453
16
- nkululeko/modelrunner.py,sha256=GuYsmUGSmJ0QxXxR8k0TZ47IDMtdGIMbm5nMq4Ix6tU,9335
16
+ nkululeko/modelrunner.py,sha256=GwDXcE2gDQXat4W0-HhHQ1BcUNCRBXMBQ4QycfHp_5c,9288
17
17
  nkululeko/multidb.py,sha256=fG3VukEWP1vreVN4gB1IRXxwwg4jLftsSEYtu0o1f78,5634
18
- nkululeko/nkuluflag.py,sha256=FCetTfgH69u4AwENgeCKVi3vBIR10Di67SfbupGQqfc,3354
18
+ nkululeko/nkuluflag.py,sha256=PGWSmZz-PiiHLgcZJAoGOI_Y-sZDVI1ksB8p5r7riWM,3725
19
19
  nkululeko/nkululeko.py,sha256=Kn3s2E3yyH8cJ7z6lkMxrnqtCxTu7-qfe9Zr_ONTD5g,1968
20
20
  nkululeko/plots.py,sha256=K88ZRPFGX_r03BT742H06Dde20xZYdltv7dxjgUiAFA,23025
21
21
  nkululeko/predict.py,sha256=sF091sSSLnEWcISx9ZcULLie3tY5XeFsQJd6b3vrxFg,2409
22
- nkululeko/reporter.py,sha256=mCy8er8z4e5hJ7XbOyy6BgZYZM6Lz-EKXHh4zlT0Zc8,12427
23
22
  nkululeko/resample.py,sha256=3WbxkwgyTe_fW38046Rjxk3knOkFdhqn2C4nfhbUurQ,2287
24
23
  nkululeko/runmanager.py,sha256=eTM1DNQKt1lxYhzt4vZyZluPXW9sWlIJHNQzex4lkJU,7624
25
24
  nkululeko/scaler.py,sha256=4nkIqoajkIkuTPK0Z02ifMN_awl6fP_i-GBYdoGYgGM,4101
26
25
  nkululeko/segment.py,sha256=YLKckX44tbvTb3LrdgYw9X4guzuF27sutl92z9DkpZU,4835
27
26
  nkululeko/syllable_nuclei.py,sha256=Sky-C__MeUDaxqHnDl2TGLLYOYvsahD35TUjWGeG31k,10047
28
- nkululeko/test.py,sha256=JRoLgqQJEhAIGetw-qlOUihSTTQ7O8DYafB0FlQESIQ,1525
29
- nkululeko/test_predictor.py,sha256=L2njNydYTX85_7RhHjfk1MN4JwE21tsBVuX4fY-XEjc,2753
27
+ nkululeko/test.py,sha256=1w624vo5KTzmFC8BUStGlLDmIEAFuJUz7J0W-gp7AxI,1677
28
+ nkululeko/test_predictor.py,sha256=_w5J8CxH6hmW3mLTKbdfmywl5QpdNAnW1Y8TE5GtlfE,3237
30
29
  nkululeko/augmenting/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
31
30
  nkululeko/augmenting/augmenter.py,sha256=XAt0dpmlnKxqyysqCgV3rcz-pRIvOz7rU7dmGDCVAzs,2905
32
31
  nkululeko/augmenting/randomsplicer.py,sha256=Z5rxdKKUpuncLWuTS6xVfVKUeVbeiYU_dLRHQ5fcg4Y,2669
@@ -95,7 +94,7 @@ nkululeko/reporting/defines.py,sha256=IsY1YgKRMaABpylVKjBJgJ5bNCEbGCVA_E6pivraqS
95
94
  nkululeko/reporting/latex_writer.py,sha256=qiCRSmB4KOD_za4oHu5x-PhwjZohzfo8wecMOwlXZwc,1886
96
95
  nkululeko/reporting/report.py,sha256=W0rcigDdjBvxZQ3pZja_gvToILYvaZ1BFtnN2qFRfYI,1060
97
96
  nkululeko/reporting/report_item.py,sha256=siWeGNgo4bAE46YBMNcsdf3jTMTy76BO9Fi6DTvDig4,533
98
- nkululeko/reporting/reporter.py,sha256=wwpY0gA-8E8d26XH3DSmXm3X0BkBw2Y0YyEiUiNU_Y0,12670
97
+ nkululeko/reporting/reporter.py,sha256=eLqwKEUTQ7v5CedzhZP2617qmXGcvi0rjyyFLOBdxtQ,12841
99
98
  nkululeko/reporting/result.py,sha256=nSN5or-Py2GPRWHkWpGRh7UCi1W0er7WLEHz8fYLk-A,742
100
99
  nkululeko/segmenting/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
101
100
  nkululeko/segmenting/seg_inaspeechsegmenter.py,sha256=pmLHuXsaqvcdYxB4PSW9l1mbQWZZBJFhi_CGabqydas,1947
@@ -104,8 +103,8 @@ nkululeko/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
104
103
  nkululeko/utils/files.py,sha256=UiGAtZRWYjHSvlmPaTMtzyNNGE6qaLaxQkybctS7iRM,4021
105
104
  nkululeko/utils/stats.py,sha256=1yUq0FTOyqkU8TwUocJRYdJaqMU5SlOBBRUun9STo2M,2829
106
105
  nkululeko/utils/util.py,sha256=_Z6OMJ3f-8TdETW9eqJYY5hwNRS5XCt9azzRnqoTTZE,12330
107
- nkululeko-0.82.4.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
108
- nkululeko-0.82.4.dist-info/METADATA,sha256=iVA5K6coH1oLy918OgAZoq6wqFXcfYIcuZBV1dOnI2g,35950
109
- nkululeko-0.82.4.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
110
- nkululeko-0.82.4.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
111
- nkululeko-0.82.4.dist-info/RECORD,,
106
+ nkululeko-0.83.1.dist-info/LICENSE,sha256=0zGP5B_W35yAcGfHPS18Q2B8UhvLRY3dQq1MhpsJU_U,1076
107
+ nkululeko-0.83.1.dist-info/METADATA,sha256=EgPYOS_ELZQmEvPWlX-klt8gmo59suFFL_HDptU474w,36080
108
+ nkululeko-0.83.1.dist-info/WHEEL,sha256=GJ7t_kWBFywbagK5eo9IoUwLW6oyOeTKmQ-9iHFVNxQ,92
109
+ nkululeko-0.83.1.dist-info/top_level.txt,sha256=DPFNNSHPjUeVKj44dVANAjuVGRCC3MusJ08lc2a8xFA,10
110
+ nkululeko-0.83.1.dist-info/RECORD,,
nkululeko/reporter.py DELETED
@@ -1,332 +0,0 @@
1
- """Reporter module.
2
-
3
- This module contains the Reporter class which is responsible for generating reports.
4
- """
5
-
6
-
7
- import ast
8
- import glob
9
- import json
10
- import math
11
-
12
- import matplotlib.pyplot as plt
13
- import numpy as np
14
- from scipy.stats import pearsonr
15
- from sklearn.metrics import (
16
- ConfusionMatrixDisplay,
17
- accuracy_score,
18
- classification_report,
19
- confusion_matrix,
20
- mean_absolute_error,
21
- mean_squared_error,
22
- r2_score,
23
- recall_score,
24
- )
25
- from sklearn.utils import resample
26
-
27
- import nkululeko.glob_conf as glob_conf
28
- from nkululeko.reporting.defines import Header
29
- from nkululeko.reporting.report_item import ReportItem
30
- from nkululeko.result import Result
31
- from nkululeko.utils.util import Util
32
-
33
-
34
- class Reporter:
35
- def __set_measure(self):
36
- if self.util.exp_is_classification():
37
- self.MEASURE = "UAR"
38
- self.result.measure = self.MEASURE
39
- self.is_classification = True
40
- else:
41
- self.is_classification = False
42
- self.measure = self.util.config_val("MODEL", "measure", "mse")
43
- if self.measure == "mse":
44
- self.MEASURE = "MSE"
45
- self.result.measure = self.MEASURE
46
- elif self.measure == "mae":
47
- self.MEASURE = "MAE"
48
- self.result.measure = self.MEASURE
49
- elif self.measure == "ccc":
50
- self.MEASURE = "CCC"
51
- self.result.measure = self.MEASURE
52
-
53
- def __init__(self, truths, preds, run, epoch):
54
- """Initialization with ground truth und predictions vector"""
55
- self.util = Util("reporter")
56
- self.format = self.util.config_val("PLOT", "format", "png")
57
- self.truths = truths
58
- self.preds = preds
59
- self.result = Result(0, 0, 0, 0, "unknown")
60
- self.run = run
61
- self.epoch = epoch
62
- self.__set_measure()
63
- self.cont_to_cat = False
64
- if len(self.truths) > 0 and len(self.preds) > 0:
65
- if self.util.exp_is_classification():
66
- self.result.test = recall_score(
67
- self.truths, self.preds, average="macro"
68
- )
69
- self.result.loss = 1 - accuracy_score(self.truths, self.preds)
70
- else:
71
- # regression experiment
72
- if self.measure == "mse":
73
- self.result.test = mean_squared_error(
74
- self.truths, self.preds)
75
- elif self.measure == "mae":
76
- self.result.test = mean_absolute_error(
77
- self.truths, self.preds)
78
- elif self.measure == "ccc":
79
- self.result.test = self.ccc(self.truths, self.preds)
80
- if math.isnan(self.result.test):
81
- self.util.debug(f"Truth: {self.truths}")
82
- self.util.debug(f"Predict.: {self.preds}")
83
- self.util.debug(f"Result is NAN: setting to -1")
84
- self.result.test = -1
85
- else:
86
- self.util.error(f"unknown measure: {self.measure}")
87
-
88
- # train and loss are being set by the model
89
-
90
- def set_id(self, run, epoch):
91
- """Make the report identifiable with run and epoch index"""
92
- self.run = run
93
- self.epoch = epoch
94
-
95
- def continuous_to_categorical(self):
96
- if self.cont_to_cat:
97
- return
98
- self.cont_to_cat = True
99
- bins = ast.literal_eval(glob_conf.config["DATA"]["bins"])
100
- self.truths = np.digitize(self.truths, bins) - 1
101
- self.preds = np.digitize(self.preds, bins) - 1
102
-
103
- def plot_confmatrix(self, plot_name, epoch):
104
- if not self.util.exp_is_classification():
105
- self.continuous_to_categorical()
106
- self._plot_confmat(self.truths, self.preds, plot_name, epoch)
107
-
108
-
109
- def plot_per_speaker(self, result_df, plot_name, function):
110
- """Plot a confusion matrix with the mode category per speakers.
111
-
112
- This function creates a confusion matrix for each speaker in the result_df.
113
- The result_df should contain the columns: preds, truths and speaker.
114
-
115
- Args:
116
- * result_df: a pandas dataframe with columns: preds, truths and speaker
117
- * plot_name: a string with the name of the plot
118
- * function: a string with the function to use for each speaker,
119
- can be 'mode' or 'mean'
120
-
121
- Returns:
122
- * None
123
- """
124
- # Initialize empty arrays for predictions and truths
125
- pred = np.zeros(0)
126
- truth = np.zeros(0)
127
-
128
- # Iterate over each speaker
129
- for s in result_df.speaker.unique():
130
- # Filter the dataframe for the current speaker
131
- s_df = result_df[result_df.speaker == s]
132
-
133
- # Get the mode or mean prediction for the current speaker
134
- mode = s_df.pred.mode().iloc[-1]
135
- mean = s_df.pred.mean()
136
- if function == "mode":
137
- s_df.pred = mode
138
- elif function == "mean":
139
- s_df.pred = mean
140
- else:
141
- self.util.error(f"unknown function {function}")
142
-
143
- # Append the current speaker's predictions and truths to the arrays
144
- pred = np.append(pred, s_df.pred.values)
145
- truth = np.append(truth, s_df["truth"].values)
146
-
147
- # If the experiment is not a classification or continuous to categorical conversion was performed,
148
- # convert the truths and predictions to categorical
149
- if not (self.is_classification or self.cont_to_cat):
150
- bins = ast.literal_eval(glob_conf.config["DATA"]["bins"])
151
- truth = np.digitize(truth, bins) - 1
152
- pred = np.digitize(pred, bins) - 1
153
-
154
- # Plot the confusion matrix for the speakers
155
- self._plot_confmat(truth, pred.astype("int"), plot_name, 0)
156
-
157
- def _plot_confmat(self, truths, preds, plot_name, epoch):
158
- # print(truths)
159
- # print(preds)
160
- fig_dir = self.util.get_path("fig_dir")
161
- labels = glob_conf.labels
162
- fig = plt.figure() # figsize=[5, 5]
163
- uar = recall_score(truths, preds, average="macro")
164
- acc = accuracy_score(truths, preds)
165
- cm = confusion_matrix(
166
- truths, preds, normalize=None
167
- ) # normalize must be one of {'true', 'pred', 'all', None}
168
- if cm.shape[0] != len(labels):
169
- self.util.error(
170
- f"mismatch between confmatrix dim ({cm.shape[0]}) and labels"
171
- f" length ({len(labels)}: {labels})"
172
- )
173
- try:
174
- disp = ConfusionMatrixDisplay(
175
- confusion_matrix=cm, display_labels=labels
176
- ).plot(cmap="Blues")
177
- except ValueError:
178
- disp = ConfusionMatrixDisplay(
179
- confusion_matrix=cm,
180
- display_labels=list(labels).remove("neutral"),
181
- ).plot(cmap="Blues")
182
-
183
- reg_res = ""
184
- if not self.is_classification:
185
- reg_res = f", {self.MEASURE}: {self.result.test:.3f}"
186
-
187
- if epoch != 0:
188
- plt.title(
189
- f"Confusion Matrix, UAR: {uar:.3f}{reg_res}, Epoch: {epoch}")
190
- else:
191
- plt.title(f"Confusion Matrix, UAR: {uar:.3f}{reg_res}")
192
- img_path = f"{fig_dir}{plot_name}.{self.format}"
193
- plt.savefig(img_path)
194
- fig.clear()
195
- plt.close(fig)
196
- plt.savefig(img_path)
197
- plt.close(fig)
198
- glob_conf.report.add_item(
199
- ReportItem(
200
- Header.HEADER_RESULTS,
201
- self.util.get_model_description(),
202
- "Confusion matrix",
203
- img_path,
204
- )
205
- )
206
-
207
- res_dir = self.util.get_path("res_dir")
208
- uar = int(uar * 1000) / 1000.0
209
- acc = int(acc * 1000) / 1000.0
210
- rpt = f"epoch: {epoch}, UAR: {uar}, ACC: {acc}"
211
- # print(rpt)
212
- self.util.debug(rpt)
213
- file_name = f"{res_dir}{self.util.get_exp_name()}_conf.txt"
214
- with open(file_name, "w") as text_file:
215
- text_file.write(rpt)
216
-
217
- def print_results(self, epoch):
218
- """Print all evaluation values to text file"""
219
- res_dir = self.util.get_path("res_dir")
220
- file_name = f"{res_dir}{self.util.get_exp_name()}_{epoch}.txt"
221
- if self.util.exp_is_classification():
222
- labels = glob_conf.labels
223
- try:
224
- rpt = classification_report(
225
- self.truths,
226
- self.preds,
227
- target_names=labels,
228
- output_dict=True,
229
- )
230
- except ValueError as e:
231
- self.util.debug(
232
- "Reporter: caught a ValueError when trying to get"
233
- " classification_report: " + e
234
- )
235
- rpt = self.result.to_string()
236
- with open(file_name, "w") as text_file:
237
- c_ress = list(range(len(labels)))
238
- for i, l in enumerate(labels):
239
- c_res = rpt[l]["f1-score"]
240
- c_ress[i] = float(f"{c_res:.3f}")
241
- self.util.debug(f"labels: {labels}")
242
- f1_per_class = f"result per class (F1 score): {c_ress}"
243
- self.util.debug(f1_per_class)
244
- rpt_str = f"{json.dumps(rpt)}\n{f1_per_class}"
245
- text_file.write(rpt_str)
246
- glob_conf.report.add_item(
247
- ReportItem(
248
- Header.HEADER_RESULTS,
249
- f"Classification result {self.util.get_model_description()}",
250
- rpt_str,
251
- )
252
- )
253
-
254
- else: # regression
255
- result = self.result.test
256
- r2 = r2_score(self.truths, self.preds)
257
- pcc = pearsonr(self.truths, self.preds)[0]
258
- measure = self.util.config_val("MODEL", "measure", "mse")
259
- with open(file_name, "w") as text_file:
260
- text_file.write(
261
- f"{measure}: {result:.3f}, r_2: {r2:.3f}, pcc {pcc:.3f}"
262
- )
263
-
264
- def make_conf_animation(self, out_name):
265
- import imageio
266
-
267
- fig_dir = self.util.get_path("fig_dir")
268
- filenames = glob.glob(
269
- fig_dir + f"{self.util.get_plot_name()}*_?_???_cnf.png")
270
- images = []
271
- for filename in filenames:
272
- images.append(imageio.imread(filename))
273
- fps = self.util.config_val("PLOT", "fps", "1")
274
- try:
275
- imageio.mimsave(fig_dir + out_name, images, fps=int(fps))
276
- except RuntimeError as e:
277
- self.util.error("error writing anim gif: " + e)
278
-
279
- def get_result(self):
280
- return self.result
281
-
282
- def plot_epoch_progression(self, reports, out_name):
283
- fig_dir = self.util.get_path("fig_dir")
284
- results, losses, train_results, losses_eval = [], [], [], []
285
- for r in reports:
286
- results.append(r.get_result().test)
287
- losses.append(r.get_result().loss)
288
- train_results.append(r.get_result().train)
289
- losses_eval.append(r.get_result().loss_eval)
290
-
291
- # do a plot per run
292
- # scale the losses so they fit on the picture
293
- losses, results, train_results, losses_eval = (
294
- np.asarray(losses),
295
- np.asarray(results),
296
- np.asarray(train_results),
297
- np.asarray(losses_eval),
298
- )
299
-
300
- if np.all((results > 1)):
301
- # scale down values
302
- results = results / 100.0
303
- train_results = train_results / 100.0
304
- # if np.all((losses < 1)):
305
- # scale up values
306
- plt.figure(dpi=200)
307
- plt.plot(train_results, "green", label="train set")
308
- plt.plot(results, "red", label="dev set")
309
- plt.plot(losses, "black", label="losses")
310
- plt.plot(losses_eval, "grey", label="losses_eval")
311
- plt.xlabel("epochs")
312
- plt.ylabel(f"{self.MEASURE}")
313
- plt.legend()
314
- plt.savefig(f"{fig_dir}{out_name}.{self.format}")
315
- plt.close()
316
-
317
- @staticmethod
318
- def ccc(ground_truth, prediction):
319
- mean_gt = np.mean(ground_truth, 0)
320
- mean_pred = np.mean(prediction, 0)
321
- var_gt = np.var(ground_truth, 0)
322
- var_pred = np.var(prediction, 0)
323
- v_pred = prediction - mean_pred
324
- v_gt = ground_truth - mean_gt
325
- cor = sum(v_pred * v_gt) / \
326
- (np.sqrt(sum(v_pred**2)) * np.sqrt(sum(v_gt**2)))
327
- sd_gt = np.std(ground_truth)
328
- sd_pred = np.std(prediction)
329
- numerator = 2 * cor * sd_gt * sd_pred
330
- denominator = var_gt + var_pred + (mean_gt - mean_pred) ** 2
331
- ccc = numerator / denominator
332
- return ccc