Spaces:
Runtime error
Runtime error
chaitanya9
commited on
Commit
·
9a38ba2
1
Parent(s):
6609b72
Upload emotion_recognition.py
Browse files- emotion_recognition.py +497 -0
emotion_recognition.py
ADDED
@@ -0,0 +1,497 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from data_extractor import load_data
|
2 |
+
from utils import extract_feature, AVAILABLE_EMOTIONS
|
3 |
+
from create_csv import write_emodb_csv, write_tess_ravdess_csv, write_custom_csv
|
4 |
+
|
5 |
+
from sklearn.metrics import accuracy_score, make_scorer, fbeta_score, mean_squared_error, mean_absolute_error
|
6 |
+
from sklearn.metrics import confusion_matrix
|
7 |
+
from sklearn.model_selection import GridSearchCV
|
8 |
+
|
9 |
+
import matplotlib.pyplot as pl
|
10 |
+
from time import time
|
11 |
+
from utils import get_best_estimators, get_audio_config
|
12 |
+
import numpy as np
|
13 |
+
import tqdm
|
14 |
+
import os
|
15 |
+
import random
|
16 |
+
import pandas as pd
|
17 |
+
|
18 |
+
|
19 |
+
class EmotionRecognizer:
|
20 |
+
"""A class for training, testing and predicting emotions based on
|
21 |
+
speech's features that are extracted and fed into `sklearn` or `keras` model"""
|
22 |
+
def __init__(self, model=None, **kwargs):
|
23 |
+
"""
|
24 |
+
Params:
|
25 |
+
model (sklearn model): the model used to detect emotions. If `model` is None, then self.determine_best_model()
|
26 |
+
will be automatically called
|
27 |
+
emotions (list): list of emotions to be used. Note that these emotions must be available in
|
28 |
+
RAVDESS_TESS & EMODB Datasets, available nine emotions are the following:
|
29 |
+
'neutral', 'calm', 'happy', 'sad', 'angry', 'fear', 'disgust', 'ps' ( pleasant surprised ), 'boredom'.
|
30 |
+
Default is ["sad", "neutral", "happy"].
|
31 |
+
tess_ravdess (bool): whether to use TESS & RAVDESS Speech datasets, default is True
|
32 |
+
emodb (bool): whether to use EMO-DB Speech dataset, default is True,
|
33 |
+
custom_db (bool): whether to use custom Speech dataset that is located in `data/train-custom`
|
34 |
+
and `data/test-custom`, default is True
|
35 |
+
tess_ravdess_name (str): the name of the output CSV file for TESS&RAVDESS dataset, default is "tess_ravdess.csv"
|
36 |
+
emodb_name (str): the name of the output CSV file for EMO-DB dataset, default is "emodb.csv"
|
37 |
+
custom_db_name (str): the name of the output CSV file for the custom dataset, default is "custom.csv"
|
38 |
+
features (list): list of speech features to use, default is ["mfcc", "chroma", "mel"]
|
39 |
+
(i.e MFCC, Chroma and MEL spectrogram )
|
40 |
+
classification (bool): whether to use classification or regression, default is True
|
41 |
+
balance (bool): whether to balance the dataset ( both training and testing ), default is True
|
42 |
+
verbose (bool/int): whether to print messages on certain tasks, default is 1
|
43 |
+
Note that when `tess_ravdess`, `emodb` and `custom_db` are set to `False`, `tess_ravdess` will be set to True
|
44 |
+
automatically.
|
45 |
+
"""
|
46 |
+
# emotions
|
47 |
+
self.emotions = kwargs.get("emotions", ["sad", "neutral", "happy"])
|
48 |
+
# make sure that there are only available emotions
|
49 |
+
self._verify_emotions()
|
50 |
+
# audio config
|
51 |
+
self.features = kwargs.get("features", ["mfcc", "chroma", "mel"])
|
52 |
+
self.audio_config = get_audio_config(self.features)
|
53 |
+
# datasets
|
54 |
+
self.tess_ravdess = kwargs.get("tess_ravdess", True)
|
55 |
+
self.emodb = kwargs.get("emodb", True)
|
56 |
+
self.custom_db = kwargs.get("custom_db", True)
|
57 |
+
|
58 |
+
if not self.tess_ravdess and not self.emodb and not self.custom_db:
|
59 |
+
self.tess_ravdess = True
|
60 |
+
|
61 |
+
self.classification = kwargs.get("classification", True)
|
62 |
+
self.balance = kwargs.get("balance", True)
|
63 |
+
self.override_csv = kwargs.get("override_csv", True)
|
64 |
+
self.verbose = kwargs.get("verbose", 1)
|
65 |
+
|
66 |
+
self.tess_ravdess_name = kwargs.get("tess_ravdess_name", "tess_ravdess.csv")
|
67 |
+
self.emodb_name = kwargs.get("emodb_name", "emodb.csv")
|
68 |
+
self.custom_db_name = kwargs.get("custom_db_name", "custom.csv")
|
69 |
+
|
70 |
+
self.verbose = kwargs.get("verbose", 1)
|
71 |
+
|
72 |
+
# set metadata path file names
|
73 |
+
self._set_metadata_filenames()
|
74 |
+
# write csv's anyway
|
75 |
+
self.write_csv()
|
76 |
+
|
77 |
+
# boolean attributes
|
78 |
+
self.data_loaded = False
|
79 |
+
self.model_trained = False
|
80 |
+
|
81 |
+
# model
|
82 |
+
if not model:
|
83 |
+
self.determine_best_model()
|
84 |
+
else:
|
85 |
+
self.model = model
|
86 |
+
|
87 |
+
def _set_metadata_filenames(self):
|
88 |
+
"""
|
89 |
+
Protected method to get all CSV (metadata) filenames into two instance attributes:
|
90 |
+
- `self.train_desc_files` for training CSVs
|
91 |
+
- `self.test_desc_files` for testing CSVs
|
92 |
+
"""
|
93 |
+
train_desc_files, test_desc_files = [], []
|
94 |
+
if self.tess_ravdess:
|
95 |
+
train_desc_files.append(f"train_{self.tess_ravdess_name}")
|
96 |
+
test_desc_files.append(f"test_{self.tess_ravdess_name}")
|
97 |
+
if self.emodb:
|
98 |
+
train_desc_files.append(f"train_{self.emodb_name}")
|
99 |
+
test_desc_files.append(f"test_{self.emodb_name}")
|
100 |
+
if self.custom_db:
|
101 |
+
train_desc_files.append(f"train_{self.custom_db_name}")
|
102 |
+
test_desc_files.append(f"test_{self.custom_db_name}")
|
103 |
+
|
104 |
+
# set them to be object attributes
|
105 |
+
self.train_desc_files = train_desc_files
|
106 |
+
self.test_desc_files = test_desc_files
|
107 |
+
|
108 |
+
def _verify_emotions(self):
|
109 |
+
"""
|
110 |
+
This method makes sure that emotions passed in parameters are valid.
|
111 |
+
"""
|
112 |
+
for emotion in self.emotions:
|
113 |
+
assert emotion in AVAILABLE_EMOTIONS, "Emotion not recognized."
|
114 |
+
|
115 |
+
def get_best_estimators(self):
|
116 |
+
"""Loads estimators from grid files and returns them"""
|
117 |
+
return get_best_estimators(self.classification)
|
118 |
+
|
119 |
+
def write_csv(self):
|
120 |
+
"""
|
121 |
+
Write available CSV files in `self.train_desc_files` and `self.test_desc_files`
|
122 |
+
determined by `self._set_metadata_filenames()` method.
|
123 |
+
"""
|
124 |
+
for train_csv_file, test_csv_file in zip(self.train_desc_files, self.test_desc_files):
|
125 |
+
# not safe approach
|
126 |
+
if os.path.isfile(train_csv_file) and os.path.isfile(test_csv_file):
|
127 |
+
# file already exists, just skip writing csv files
|
128 |
+
if not self.override_csv:
|
129 |
+
continue
|
130 |
+
if self.emodb_name in train_csv_file:
|
131 |
+
write_emodb_csv(self.emotions, train_name=train_csv_file, test_name=test_csv_file, verbose=self.verbose)
|
132 |
+
if self.verbose:
|
133 |
+
print("[+] Writed EMO-DB CSV File")
|
134 |
+
elif self.tess_ravdess_name in train_csv_file:
|
135 |
+
write_tess_ravdess_csv(self.emotions, train_name=train_csv_file, test_name=test_csv_file, verbose=self.verbose)
|
136 |
+
if self.verbose:
|
137 |
+
print("[+] Writed TESS & RAVDESS DB CSV File")
|
138 |
+
elif self.custom_db_name in train_csv_file:
|
139 |
+
write_custom_csv(emotions=self.emotions, train_name=train_csv_file, test_name=test_csv_file, verbose=self.verbose)
|
140 |
+
if self.verbose:
|
141 |
+
print("[+] Writed Custom DB CSV File")
|
142 |
+
|
143 |
+
def load_data(self):
|
144 |
+
"""
|
145 |
+
Loads and extracts features from the audio files for the db's specified
|
146 |
+
"""
|
147 |
+
if not self.data_loaded:
|
148 |
+
result = load_data(self.train_desc_files, self.test_desc_files, self.audio_config, self.classification,
|
149 |
+
emotions=self.emotions, balance=self.balance)
|
150 |
+
self.X_train = result['X_train']
|
151 |
+
self.X_test = result['X_test']
|
152 |
+
self.y_train = result['y_train']
|
153 |
+
self.y_test = result['y_test']
|
154 |
+
self.train_audio_paths = result['train_audio_paths']
|
155 |
+
self.test_audio_paths = result['test_audio_paths']
|
156 |
+
self.balance = result["balance"]
|
157 |
+
if self.verbose:
|
158 |
+
print("[+] Data loaded")
|
159 |
+
self.data_loaded = True
|
160 |
+
|
161 |
+
def train(self, verbose=1):
|
162 |
+
"""
|
163 |
+
Train the model, if data isn't loaded, it 'll be loaded automatically
|
164 |
+
"""
|
165 |
+
if not self.data_loaded:
|
166 |
+
# if data isn't loaded yet, load it then
|
167 |
+
self.load_data()
|
168 |
+
if not self.model_trained:
|
169 |
+
self.model.fit(X=self.X_train, y=self.y_train)
|
170 |
+
self.model_trained = True
|
171 |
+
if verbose:
|
172 |
+
print("[+] Model trained")
|
173 |
+
|
174 |
+
def predict(self, audio_path):
|
175 |
+
"""
|
176 |
+
given an `audio_path`, this method extracts the features
|
177 |
+
and predicts the emotion
|
178 |
+
"""
|
179 |
+
feature = extract_feature(audio_path, **self.audio_config).reshape(1, -1)
|
180 |
+
return self.model.predict(feature)[0]
|
181 |
+
|
182 |
+
def predict_proba(self, audio_path):
|
183 |
+
"""
|
184 |
+
Predicts the probability of each emotion.
|
185 |
+
"""
|
186 |
+
if self.classification:
|
187 |
+
feature = extract_feature(audio_path, **self.audio_config).reshape(1, -1)
|
188 |
+
proba = self.model.predict_proba(feature)[0]
|
189 |
+
result = {}
|
190 |
+
for emotion, prob in zip(self.model.classes_, proba):
|
191 |
+
result[emotion] = prob
|
192 |
+
return result
|
193 |
+
else:
|
194 |
+
raise NotImplementedError("Probability prediction doesn't make sense for regression")
|
195 |
+
|
196 |
+
def grid_search(self, params, n_jobs=2, verbose=1):
|
197 |
+
"""
|
198 |
+
Performs GridSearchCV on `params` passed on the `self.model`
|
199 |
+
And returns the tuple: (best_estimator, best_params, best_score).
|
200 |
+
"""
|
201 |
+
score = accuracy_score if self.classification else mean_absolute_error
|
202 |
+
grid = GridSearchCV(estimator=self.model, param_grid=params, scoring=make_scorer(score),
|
203 |
+
n_jobs=n_jobs, verbose=verbose, cv=3)
|
204 |
+
grid_result = grid.fit(self.X_train, self.y_train)
|
205 |
+
return grid_result.best_estimator_, grid_result.best_params_, grid_result.best_score_
|
206 |
+
|
207 |
+
def determine_best_model(self):
|
208 |
+
"""
|
209 |
+
Loads best estimators and determine which is best for test data,
|
210 |
+
and then set it to `self.model`.
|
211 |
+
In case of regression, the metric used is MSE and accuracy for classification.
|
212 |
+
Note that the execution of this method may take several minutes due
|
213 |
+
to training all estimators (stored in `grid` folder) for determining the best possible one.
|
214 |
+
"""
|
215 |
+
if not self.data_loaded:
|
216 |
+
self.load_data()
|
217 |
+
|
218 |
+
# loads estimators
|
219 |
+
estimators = self.get_best_estimators()
|
220 |
+
|
221 |
+
result = []
|
222 |
+
|
223 |
+
if self.verbose:
|
224 |
+
estimators = tqdm.tqdm(estimators)
|
225 |
+
|
226 |
+
for estimator, params, cv_score in estimators:
|
227 |
+
if self.verbose:
|
228 |
+
estimators.set_description(f"Evaluating {estimator.__class__.__name__}")
|
229 |
+
detector = EmotionRecognizer(estimator, emotions=self.emotions, tess_ravdess=self.tess_ravdess,
|
230 |
+
emodb=self.emodb, custom_db=self.custom_db, classification=self.classification,
|
231 |
+
features=self.features, balance=self.balance, override_csv=False)
|
232 |
+
# data already loaded
|
233 |
+
detector.X_train = self.X_train
|
234 |
+
detector.X_test = self.X_test
|
235 |
+
detector.y_train = self.y_train
|
236 |
+
detector.y_test = self.y_test
|
237 |
+
detector.data_loaded = True
|
238 |
+
# train the model
|
239 |
+
detector.train(verbose=0)
|
240 |
+
# get test accuracy
|
241 |
+
accuracy = detector.test_score()
|
242 |
+
# append to result
|
243 |
+
result.append((detector.model, accuracy))
|
244 |
+
|
245 |
+
# sort the result
|
246 |
+
# regression: best is the lower, not the higher
|
247 |
+
# classification: best is higher, not the lower
|
248 |
+
result = sorted(result, key=lambda item: item[1], reverse=self.classification)
|
249 |
+
best_estimator = result[0][0]
|
250 |
+
accuracy = result[0][1]
|
251 |
+
self.model = best_estimator
|
252 |
+
self.model_trained = True
|
253 |
+
if self.verbose:
|
254 |
+
if self.classification:
|
255 |
+
print(f"[+] Best model determined: {self.model.__class__.__name__} with {accuracy*100:.3f}% test accuracy")
|
256 |
+
else:
|
257 |
+
print(f"[+] Best model determined: {self.model.__class__.__name__} with {accuracy:.5f} mean absolute error")
|
258 |
+
|
259 |
+
def test_score(self):
|
260 |
+
"""
|
261 |
+
Calculates score on testing data
|
262 |
+
if `self.classification` is True, the metric used is accuracy,
|
263 |
+
Mean-Squared-Error is used otherwise (regression)
|
264 |
+
"""
|
265 |
+
y_pred = self.model.predict(self.X_test)
|
266 |
+
if self.classification:
|
267 |
+
return accuracy_score(y_true=self.y_test, y_pred=y_pred)
|
268 |
+
else:
|
269 |
+
return mean_squared_error(y_true=self.y_test, y_pred=y_pred)
|
270 |
+
|
271 |
+
def train_score(self):
|
272 |
+
"""
|
273 |
+
Calculates accuracy score on training data
|
274 |
+
if `self.classification` is True, the metric used is accuracy,
|
275 |
+
Mean-Squared-Error is used otherwise (regression)
|
276 |
+
"""
|
277 |
+
y_pred = self.model.predict(self.X_train)
|
278 |
+
if self.classification:
|
279 |
+
return accuracy_score(y_true=self.y_train, y_pred=y_pred)
|
280 |
+
else:
|
281 |
+
return mean_squared_error(y_true=self.y_train, y_pred=y_pred)
|
282 |
+
|
283 |
+
def train_fbeta_score(self, beta):
|
284 |
+
y_pred = self.model.predict(self.X_train)
|
285 |
+
return fbeta_score(self.y_train, y_pred, beta, average='micro')
|
286 |
+
|
287 |
+
def test_fbeta_score(self, beta):
|
288 |
+
y_pred = self.model.predict(self.X_test)
|
289 |
+
return fbeta_score(self.y_test, y_pred, beta, average='micro')
|
290 |
+
|
291 |
+
def confusion_matrix(self, percentage=True, labeled=True):
|
292 |
+
"""
|
293 |
+
Computes confusion matrix to evaluate the test accuracy of the classification
|
294 |
+
and returns it as numpy matrix or pandas dataframe (depends on params).
|
295 |
+
params:
|
296 |
+
percentage (bool): whether to use percentage instead of number of samples, default is True.
|
297 |
+
labeled (bool): whether to label the columns and indexes in the dataframe.
|
298 |
+
"""
|
299 |
+
if not self.classification:
|
300 |
+
raise NotImplementedError("Confusion matrix works only when it is a classification problem")
|
301 |
+
y_pred = self.model.predict(self.X_test)
|
302 |
+
matrix = confusion_matrix(self.y_test, y_pred, labels=self.emotions).astype(np.float32)
|
303 |
+
if percentage:
|
304 |
+
for i in range(len(matrix)):
|
305 |
+
matrix[i] = matrix[i] / np.sum(matrix[i])
|
306 |
+
# make it percentage
|
307 |
+
matrix *= 100
|
308 |
+
if labeled:
|
309 |
+
matrix = pd.DataFrame(matrix, index=[ f"true_{e}" for e in self.emotions ],
|
310 |
+
columns=[ f"predicted_{e}" for e in self.emotions ])
|
311 |
+
return matrix
|
312 |
+
|
313 |
+
def draw_confusion_matrix(self):
|
314 |
+
"""Calculates the confusion matrix and shows it"""
|
315 |
+
matrix = self.confusion_matrix(percentage=False, labeled=False)
|
316 |
+
#TODO: add labels, title, legends, etc.
|
317 |
+
pl.imshow(matrix, cmap="binary")
|
318 |
+
pl.show()
|
319 |
+
|
320 |
+
def get_n_samples(self, emotion, partition):
|
321 |
+
"""Returns number data samples of the `emotion` class in a particular `partition`
|
322 |
+
('test' or 'train')
|
323 |
+
"""
|
324 |
+
if partition == "test":
|
325 |
+
return len([y for y in self.y_test if y == emotion])
|
326 |
+
elif partition == "train":
|
327 |
+
return len([y for y in self.y_train if y == emotion])
|
328 |
+
|
329 |
+
def get_samples_by_class(self):
|
330 |
+
"""
|
331 |
+
Returns a dataframe that contains the number of training
|
332 |
+
and testing samples for all emotions.
|
333 |
+
Note that if data isn't loaded yet, it'll be loaded
|
334 |
+
"""
|
335 |
+
if not self.data_loaded:
|
336 |
+
self.load_data()
|
337 |
+
train_samples = []
|
338 |
+
test_samples = []
|
339 |
+
total = []
|
340 |
+
for emotion in self.emotions:
|
341 |
+
n_train = self.get_n_samples(emotion, "train")
|
342 |
+
n_test = self.get_n_samples(emotion, "test")
|
343 |
+
train_samples.append(n_train)
|
344 |
+
test_samples.append(n_test)
|
345 |
+
total.append(n_train + n_test)
|
346 |
+
|
347 |
+
# get total
|
348 |
+
total.append(sum(train_samples) + sum(test_samples))
|
349 |
+
train_samples.append(sum(train_samples))
|
350 |
+
test_samples.append(sum(test_samples))
|
351 |
+
return pd.DataFrame(data={"train": train_samples, "test": test_samples, "total": total}, index=self.emotions + ["total"])
|
352 |
+
|
353 |
+
def get_random_emotion(self, emotion, partition="train"):
|
354 |
+
"""
|
355 |
+
Returns random `emotion` data sample index on `partition`.
|
356 |
+
"""
|
357 |
+
if partition == "train":
|
358 |
+
index = random.choice(list(range(len(self.y_train))))
|
359 |
+
while self.y_train[index] != emotion:
|
360 |
+
index = random.choice(list(range(len(self.y_train))))
|
361 |
+
elif partition == "test":
|
362 |
+
index = random.choice(list(range(len(self.y_test))))
|
363 |
+
while self.y_train[index] != emotion:
|
364 |
+
index = random.choice(list(range(len(self.y_test))))
|
365 |
+
else:
|
366 |
+
raise TypeError("Unknown partition, only 'train' or 'test' is accepted")
|
367 |
+
|
368 |
+
return index
|
369 |
+
|
370 |
+
|
371 |
+
def plot_histograms(classifiers=True, beta=0.5, n_classes=3, verbose=1):
|
372 |
+
"""
|
373 |
+
Loads different estimators from `grid` folder and calculate some statistics to plot histograms.
|
374 |
+
Params:
|
375 |
+
classifiers (bool): if `True`, this will plot classifiers, regressors otherwise.
|
376 |
+
beta (float): beta value for calculating fbeta score for various estimators.
|
377 |
+
n_classes (int): number of classes
|
378 |
+
"""
|
379 |
+
# get the estimators from the performed grid search result
|
380 |
+
estimators = get_best_estimators(classifiers)
|
381 |
+
|
382 |
+
final_result = {}
|
383 |
+
for estimator, params, cv_score in estimators:
|
384 |
+
final_result[estimator.__class__.__name__] = []
|
385 |
+
for i in range(3):
|
386 |
+
result = {}
|
387 |
+
# initialize the class
|
388 |
+
detector = EmotionRecognizer(estimator, verbose=0)
|
389 |
+
# load the data
|
390 |
+
detector.load_data()
|
391 |
+
if i == 0:
|
392 |
+
# first get 1% of sample data
|
393 |
+
sample_size = 0.01
|
394 |
+
elif i == 1:
|
395 |
+
# second get 10% of sample data
|
396 |
+
sample_size = 0.1
|
397 |
+
elif i == 2:
|
398 |
+
# last get all the data
|
399 |
+
sample_size = 1
|
400 |
+
# calculate number of training and testing samples
|
401 |
+
n_train_samples = int(len(detector.X_train) * sample_size)
|
402 |
+
n_test_samples = int(len(detector.X_test) * sample_size)
|
403 |
+
# set the data
|
404 |
+
detector.X_train = detector.X_train[:n_train_samples]
|
405 |
+
detector.X_test = detector.X_test[:n_test_samples]
|
406 |
+
detector.y_train = detector.y_train[:n_train_samples]
|
407 |
+
detector.y_test = detector.y_test[:n_test_samples]
|
408 |
+
# calculate train time
|
409 |
+
t_train = time()
|
410 |
+
detector.train()
|
411 |
+
t_train = time() - t_train
|
412 |
+
# calculate test time
|
413 |
+
t_test = time()
|
414 |
+
test_accuracy = detector.test_score()
|
415 |
+
t_test = time() - t_test
|
416 |
+
# set the result to the dictionary
|
417 |
+
result['train_time'] = t_train
|
418 |
+
result['pred_time'] = t_test
|
419 |
+
result['acc_train'] = cv_score
|
420 |
+
result['acc_test'] = test_accuracy
|
421 |
+
result['f_train'] = detector.train_fbeta_score(beta)
|
422 |
+
result['f_test'] = detector.test_fbeta_score(beta)
|
423 |
+
if verbose:
|
424 |
+
print(f"[+] {estimator.__class__.__name__} with {sample_size*100}% ({n_train_samples}) data samples achieved {cv_score*100:.3f}% Validation Score in {t_train:.3f}s & {test_accuracy*100:.3f}% Test Score in {t_test:.3f}s")
|
425 |
+
# append the dictionary to the list of results
|
426 |
+
final_result[estimator.__class__.__name__].append(result)
|
427 |
+
if verbose:
|
428 |
+
print()
|
429 |
+
visualize(final_result, n_classes=n_classes)
|
430 |
+
|
431 |
+
|
432 |
+
|
433 |
+
def visualize(results, n_classes):
|
434 |
+
"""
|
435 |
+
Visualization code to display results of various learners.
|
436 |
+
|
437 |
+
inputs:
|
438 |
+
- results: a dictionary of lists of dictionaries that contain various results on the corresponding estimator
|
439 |
+
- n_classes: number of classes
|
440 |
+
"""
|
441 |
+
|
442 |
+
n_estimators = len(results)
|
443 |
+
|
444 |
+
# naive predictor
|
445 |
+
accuracy = 1 / n_classes
|
446 |
+
f1 = 1 / n_classes
|
447 |
+
# Create figure
|
448 |
+
fig, ax = pl.subplots(2, 4, figsize = (11,7))
|
449 |
+
# Constants
|
450 |
+
bar_width = 0.4
|
451 |
+
colors = [ (random.random(), random.random(), random.random()) for _ in range(n_estimators) ]
|
452 |
+
# Super loop to plot four panels of data
|
453 |
+
for k, learner in enumerate(results.keys()):
|
454 |
+
for j, metric in enumerate(['train_time', 'acc_train', 'f_train', 'pred_time', 'acc_test', 'f_test']):
|
455 |
+
for i in np.arange(3):
|
456 |
+
x = bar_width * n_estimators
|
457 |
+
# Creative plot code
|
458 |
+
ax[j//3, j%3].bar(i*x+k*(bar_width), results[learner][i][metric], width = bar_width, color = colors[k])
|
459 |
+
ax[j//3, j%3].set_xticks([x-0.2, x*2-0.2, x*3-0.2])
|
460 |
+
ax[j//3, j%3].set_xticklabels(["1%", "10%", "100%"])
|
461 |
+
ax[j//3, j%3].set_xlabel("Training Set Size")
|
462 |
+
ax[j//3, j%3].set_xlim((-0.2, x*3))
|
463 |
+
# Add unique y-labels
|
464 |
+
ax[0, 0].set_ylabel("Time (in seconds)")
|
465 |
+
ax[0, 1].set_ylabel("Accuracy Score")
|
466 |
+
ax[0, 2].set_ylabel("F-score")
|
467 |
+
ax[1, 0].set_ylabel("Time (in seconds)")
|
468 |
+
ax[1, 1].set_ylabel("Accuracy Score")
|
469 |
+
ax[1, 2].set_ylabel("F-score")
|
470 |
+
# Add titles
|
471 |
+
ax[0, 0].set_title("Model Training")
|
472 |
+
ax[0, 1].set_title("Accuracy Score on Training Subset")
|
473 |
+
ax[0, 2].set_title("F-score on Training Subset")
|
474 |
+
ax[1, 0].set_title("Model Predicting")
|
475 |
+
ax[1, 1].set_title("Accuracy Score on Testing Set")
|
476 |
+
ax[1, 2].set_title("F-score on Testing Set")
|
477 |
+
# Add horizontal lines for naive predictors
|
478 |
+
ax[0, 1].axhline(y = accuracy, xmin = -0.1, xmax = 3.0, linewidth = 1, color = 'k', linestyle = 'dashed')
|
479 |
+
ax[1, 1].axhline(y = accuracy, xmin = -0.1, xmax = 3.0, linewidth = 1, color = 'k', linestyle = 'dashed')
|
480 |
+
ax[0, 2].axhline(y = f1, xmin = -0.1, xmax = 3.0, linewidth = 1, color = 'k', linestyle = 'dashed')
|
481 |
+
ax[1, 2].axhline(y = f1, xmin = -0.1, xmax = 3.0, linewidth = 1, color = 'k', linestyle = 'dashed')
|
482 |
+
# Set y-limits for score panels
|
483 |
+
ax[0, 1].set_ylim((0, 1))
|
484 |
+
ax[0, 2].set_ylim((0, 1))
|
485 |
+
ax[1, 1].set_ylim((0, 1))
|
486 |
+
ax[1, 2].set_ylim((0, 1))
|
487 |
+
# Set additional plots invisibles
|
488 |
+
ax[0, 3].set_visible(False)
|
489 |
+
ax[1, 3].axis('off')
|
490 |
+
# Create legend
|
491 |
+
for i, learner in enumerate(results.keys()):
|
492 |
+
pl.bar(0, 0, color=colors[i], label=learner)
|
493 |
+
pl.legend()
|
494 |
+
# Aesthetics
|
495 |
+
pl.suptitle("Performance Metrics for Three Supervised Learning Models", fontsize = 16, y = 1.10)
|
496 |
+
pl.tight_layout()
|
497 |
+
pl.show()
|