def _Map_Dict_Generate(self):
self.map_Dict = {'Phone': {}, 'Feature': {}}
identifiers = ['ALL'] if hp_Dict['Hidden_Analysis']['Only_All'] else (
self.identifier_List + ['ALL'])
criteria = np.arange(
start=hp_Dict['Hidden_Analysis']['Sensitive_Index']['Criteria'][0],
stop=hp_Dict['Hidden_Analysis']['Sensitive_Index']['Criteria'][1] +
hp_Dict['Hidden_Analysis']['Sensitive_Index']['Criteria'][2],
step=hp_Dict['Hidden_Analysis']['Sensitive_Index']['Criteria'][2],
)
progress_Index = 0
for identifier in identifiers:
averages = np.stack(
[
np.mean(self.activation_Dict['Phone'][phone, identifier]
[hp_Dict['Hidden_Analysis']['Sensitive_Index']
['Step_Range'][0]:hp_Dict['Hidden_Analysis']
['Sensitive_Index']['Step_Range'][1]],
axis=(0, 1)) # [File_Nums, Time, Dim] -> [Dim]
for phone in self.phone_List
],
axis=1) # [Dim, Phone_Nums]
for criterion in criteria:
self.map_Dict['Phone'][identifier, criterion] = np.stack(
[
self._Map_Calc(x, criterion) # [Phone_Nums]
for x in averages
],
axis=1) # [Phone_Nums, Dim]
progress_Index += 1
progress(progress_Index,
len(identifiers) * len(criteria) * 2,
status='Map generating...')
for identifier in identifiers:
averages = np.stack(
[
np.mean(self.activation_Dict['Feature'][feature,
identifier],
axis=(0, 1)) # [File_Nums, Time, Dim] -> [Dim]
for feature in self.feature_List
],
axis=1) # [Dim, Feature_Nums]
for criterion in criteria:
self.map_Dict['Feature'][identifier, criterion] = np.stack(
[
self._Map_Calc(x, criterion) # [Feature_Nums]
for x in averages
],
axis=1) # [Feature_Nums, Dim]
progress_Index += 1
progress(progress_Index,
len(identifiers) * len(criteria) * 2,
status='Map generating...')
print()
def _Flow_Dict_Generate(self):
self.flow_Dict = {'Phone': {}, 'Feature': {}}
identifiers = ['ALL'] if hp_Dict['Hidden_Analysis']['Only_All'] else (
self.identifier_List + ['ALL'])
for index, identifier in enumerate(identifiers):
self.flow_Dict['Phone'][identifier] = np.stack(
[
np.transpose(
np.mean(self.activation_Dict['Phone'][phone,
identifier],
axis=0)
) # [File_Nums, Time, Dim] -> [Time, Dim] -> [Dim, Time]
for phone in self.phone_List
],
axis=1) # [Dim, Phone_Nums, Time]
self.flow_Dict['Feature'][identifier] = np.stack(
[
np.transpose(
np.mean(self.activation_Dict['Feature'][feature,
identifier],
axis=0)
) # [File_Nums, Time, Dim] -> [Time, Dim] -> [Dim, Time]
for feature in self.feature_List
],
axis=1) # [Dim, Feature_Nums, Time]
progress(index + 1, len(identifiers), status='Flow generating...')
print()
def Activation_Dict_Generate(self):
pattern_List = self.model.feeder.Get_Test_Pattern_from_Wav(
wav_Path_List=self.wav_Path_List)
hiddens = []
for index, pattern_Dict in enumerate(pattern_List):
hiddens.append(self.model.Hidden_Step(**pattern_Dict))
progress(index + 1,
len(pattern_List),
status='Activation dict generating...')
print()
hiddens = np.vstack(hiddens) # [File_Nums, Time, Dim]
self.activation_Dict = {
'Phone': { # Gathering by phoneme
(phone, identifier):
hiddens[[self.wav_Path_Index_Dict[path]
for path in path_List]] # [File_Nums, Time, Dim]
for (phone, identifier
), path_List in self.phone_File_List_Dict.items()
},
'Feature': { # Gathering by feature
(feature, identifier):
hiddens[[self.wav_Path_Index_Dict[path]
for path in path_List]] # [File_Nums, Time, Dim]
for (feature, identifier
), path_List in self.feature_File_List_Dict.items()
}
}
def Test(self, epoch):
infos, pattern_List = self.feeder.Get_Test_Pattern()
logits = []
for batch_Index, patterns in enumerate(pattern_List):
logits.append(self.Test_Step(**patterns).numpy())
progress(
batch_Index + 1,
len(pattern_List),
status='Testing'
)
print()
logits = np.vstack(logits)
export_Thread = Thread(
target=self.Export_Test,
args=(infos, logits, epoch)
) #Turning on a new thread for saving result
export_Thread.daemon = True
export_Thread.start()
return export_Thread
def body(step, samples):
current_Local_Condition = tf.expand_dims(local_Conditions[:,
step, :],
axis=1)
# global_Condition is always same. Thus, there is no step slicing.
x = self.layer_Dict['First'](inputs=tf.expand_dims(samples[:, -1],
axis=1),
training=tf.convert_to_tensor(False))
skips = 0
for block_Index in range(
hp_Dict['WaveNet']['ResConvGLU']['Blocks']):
for stack_Index in range(
hp_Dict['WaveNet']['ResConvGLU']['Stacks_in_Block']):
x, new_Skips = self.layer_Dict['ResConvGLU_{}_{}'.format(
block_Index, stack_Index
)](inputs=[x, current_Local_Condition, global_Conditions])
skips += new_Skips
skips *= np.sqrt(
1.0 / (hp_Dict['WaveNet']['ResConvGLU']['Blocks'] *
hp_Dict['WaveNet']['ResConvGLU']['Stacks_in_Block']))
logit = self.layer_Dict['Last'](skips)
samples = tf.concat(
[samples, Sample_from_Discretized_Mix_Logistic(logit)],
axis=-1)
try:
progress(step + 1,
local_Conditions.shape[1],
status='({}/{})'.format(
step + 1,
local_Conditions.get_shape()
[1])) #initial time it will be ignored.
except:
pass
return step + 1, samples
def Get_Test_Pattern(self):
pattern_Info_List = []
pattern_Info_List.extend([
(word, identifier, 'Training')
for word, identifier in self.pattern_Path_Dict['Training'].keys()
])
pattern_Info_List.extend([
(word, identifier, 'Pattern_Excluded') for word, identifier in
self.pattern_Path_Dict['Pattern_Excluded'].keys()
])
pattern_Info_List.extend([
(word, identifier, 'Identifier_Excluded') for word, identifier in
self.pattern_Path_Dict['Identifier_Excluded'].keys()
])
pattern_Info_List.extend([
(word, identifier, 'Test_Only')
for word, identifier in self.pattern_Path_Dict['Test_Only'].keys()
])
pattern_Path_List = []
pattern_Path_List.extend(
[path for path in self.pattern_Path_Dict['Training'].values()])
pattern_Path_List.extend([
path
for path in self.pattern_Path_Dict['Pattern_Excluded'].values()
])
pattern_Path_List.extend([
path
for path in self.pattern_Path_Dict['Identifier_Excluded'].values()
])
pattern_Path_List.extend(
[path for path in self.pattern_Path_Dict['Test_Only'].values()])
patterns = {}
for index, path in enumerate(pattern_Path_List):
with open(
os.path.join(hp_Dict['Pattern']['Pattern_Path'],
path).replace("\\", "/"), "rb") as f:
patterns[path] = pickle.load(f)
progress(index + 1,
len(pattern_Path_List),
status='Test pattern loading')
print()
max_Step = max(
[pattern['Acoustic'].shape[0] for pattern in patterns.values()])
pattern_Batch_List = [ # Split pattern list to genrate batchs
pattern_Path_List[x:x + hp_Dict['Train']['Batch_Size']] for x in
range(0, len(pattern_Path_List), hp_Dict['Train']['Batch_Size'])
]
test_Pattern_List = []
for pattern_Batch in pattern_Batch_List:
acoustics = []
# semantics = []
# acoustic_Steps = []
for path in pattern_Batch:
pattern_Dict = patterns[path]
acoustics.append(pattern_Dict['Acoustic'])
# semantics.append(pattern_Dict['Semantic'])
# acoustic_Steps.append(pattern_Dict['Acoustic'].shape[0])
acoustics = self.Force_Pattern_Stack(
acoustics, max_Step=max_Step).astype(np.float32)
# semantics = np.stack(semantics, axis= 0).astype(np.float32)
# acoustic_Steps = np.stack(acoustic_Steps, axis= 0).astype(np.int32)
test_Pattern_List.append({
'acoustics': acoustics,
# 'acoustic_Steps': acoustic_Steps,
# 'semantics': semantics,
})
return pattern_Info_List, test_Pattern_List
def Export_Inference(self,
letter_String_List,
pronunciation_List,
hiddens,
outputs,
word_Label_Indices,
phoneme_Label_Indices,
added_Word_Labels,
epoch,
trained_Pattern_Index_Dict,
export_Raw=False,
file_Tag='Inference'):
#Getting the count how many time each pattern is trained
trained_Pattern_Count_Dict = {
index: 0
for index in self.feeder.word_Index_Dict.values()
}
for index_List in trained_Pattern_Index_Dict.values():
for index in index_List:
trained_Pattern_Count_Dict[index] += 1
if export_Raw: #Pickled raw data saving.
export_Dict = {
'Epoch': epoch,
'Pattern_Pair':
list(zip(letter_String_List, pronunciation_List)),
'Hidden': hiddens,
'Output': outputs,
'Trained_Pattern_Count_Dict': trained_Pattern_Count_Dict
}
with open(
os.path.join(
self.export_Path, 'Inference',
'{}{}.Raw.pickle'.format(
'E_{}.'.format(epoch) if not epoch is None else '',
file_Tag)), 'wb') as f:
pickle.dump(export_Dict, f, protocol=4)
pattern_Index_List = list(range(
hiddens.shape[0])) #Getting result index
pattern_Index_Batch_List = [
pattern_Index_List[x:x + hp_Dict['Analyzer']['Batch_Size']]
for x in range(0, len(pattern_Index_List), hp_Dict['Analyzer']
['Batch_Size'])
] #Analyzing uses much memory, thus pattern must be splitted. Generating batch for analyze.
hidden_Dict = {} # A space to store analyzing results.
result_Dict = {} # A space to store analyzing results.
for index, pattern_Index_Batch in enumerate(pattern_Index_Batch_List):
batch_Hidden_Dict = self.hidden_Analyzer(
inputs=hiddens[pattern_Index_Batch]
) #In this line, hidden analyzing run. To know details, please see 'Analyzer.py'.
for key, value in batch_Hidden_Dict.items():
if not key in hidden_Dict.keys():
hidden_Dict[key] = []
hidden_Dict[key].append(value.numpy())
batch_Result_Dict = self.output_Analyzer(
inputs=outputs[pattern_Index_Batch],
word_label_indices=word_Label_Indices[pattern_Index_Batch],
phoneme_label_indices=phoneme_Label_Indices[
pattern_Index_Batch],
added_Word_Labels=added_Word_Labels
) #In this line, result analyzing run. To know details, please see 'Analyzer.py'.
for key, value in batch_Result_Dict.items():
if not key in result_Dict.keys():
result_Dict[key] = []
result_Dict[key].append(value.numpy())
progress(index + 1,
len(pattern_Index_Batch_List),
status='Inference analyzer running')
print()
hidden_Dict = {
key: np.hstack(value_List)
if len(value_List[0].shape) == 1 else np.vstack(value_List)
for key, value_List in hidden_Dict.items()
} # Stack each hidden results.
result_Dict = {
key: np.hstack(value_List)
if len(value_List[0].shape) == 1 else np.vstack(value_List)
for key, value_List in result_Dict.items()
} # Stack each hidden results.
index_Phoneme_Dict = {
index: phoneme
for phoneme, index in self.feeder.phoneme_Index_Dict.items()
if type(phoneme) == str
} # Result is a index vector. To change to phoneme list, generating index to phoneme matching.
# Result text file generating
column_Title_List = [
'Epoch',
'Ortho',
'Phono',
'Length',
'Probability',
'MeanRT',
'Trained_Count',
'Cosine_Similarity',
'Mean_Squared_Error',
'Euclidean_Distance',
'Cross_Entropy',
'Exported_Pronunciation',
'Accuracy_Max_CS',
'Accuracy_Min_MSE',
'Accuracy_Min_ED',
'Accuracy_Min_CE',
'Accuracy_Pronunciation',
'Hidden_Cosine_Similarity',
'Hidden_Mean_Squared_Error',
'Hidden_Euclidean_Distance',
'Hidden_Cross_Entropy',
]
export_List = ['\t'.join(column_Title_List)]
for index, (letter_String, target_Pronunciation, rt_CS, rt_MSE, rt_ED,
rt_CE, exported_Pronunciation, acc_CS, acc_MSE, acc_ED,
acc_CE, acc_Pronunciation, hidden_CS, hidden_MSE,
hidden_ED, hidden_CE) in enumerate(
zip(
letter_String_List,
pronunciation_List,
result_Dict['RT', 'CS'],
result_Dict['RT', 'MSE'],
result_Dict['RT', 'ED'],
result_Dict['RT', 'CE'],
result_Dict['Export', 'Pronunciation'],
result_Dict['ACC', 'Max_CS'],
result_Dict['ACC', 'Min_MSE'],
result_Dict['ACC', 'Min_ED'],
result_Dict['ACC', 'Min_CE'],
result_Dict['ACC', 'Pronunciation'],
hidden_Dict['CS'],
hidden_Dict['MSE'],
hidden_Dict['ED'],
hidden_Dict['CE'],
)):
is_Word = letter_String in self.feeder.word_Index_Dict.keys()
new_Line_List = [
str(epoch or 'None'),
letter_String,
target_Pronunciation,
str(len(letter_String)),
str(self.feeder.frequency_Dict[letter_String])
if is_Word else 'None',
str(self.feeder.human_RT_Dict[letter_String])
if is_Word else 'None',
str(trained_Pattern_Count_Dict[index]) if is_Word else 'None',
str(rt_CS),
str(rt_MSE),
str(rt_ED),
str(rt_CE),
''.join([
index_Phoneme_Dict[phoneme_Index]
for phoneme_Index in exported_Pronunciation
]),
str(acc_CS),
str(acc_MSE),
str(acc_ED),
str(acc_CE),
str(acc_Pronunciation),
str(hidden_CS),
str(hidden_MSE),
str(hidden_ED),
str(hidden_CE),
]
export_List.append('\t'.join(new_Line_List))
with open(
os.path.join(
self.export_Path, 'Inference', '{}{}.Summary.txt'.format(
'E_{}.'.format(epoch) if not epoch is None else '',
file_Tag)), 'w') as f:
f.write('\n'.join(export_List))
def RSA_Generate(self, permutation_Nums=100000):
os.makedirs(self.export_Path, exist_ok=True)
rsa_Dict = {}
permutation_Cor_List_Dict = {}
rsa_Dict['EARShot', 'Mesgarani'] = {
metric_Type:
self.RSA_Calc(self.data_Array,
self.mestarani_Distance_Dict[metric_Type],
metric_Type, False)
for metric_Type in self.mertic_Type_List
}
permutation_Cor_List_Dict['EARShot', 'Mesgarani'] = {}
for metric_Type in self.mertic_Type_List:
permutation_Cor_List_Dict['EARShot', 'Mesgarani'][metric_Type] = []
for index in range(permutation_Nums):
permutation_Cor_List_Dict[
'EARShot', 'Mesgarani'][metric_Type].append(
self.RSA_Calc(
self.data_Array,
self.mestarani_Distance_Dict[metric_Type],
metric_Type, True)[1])
progress(index + 1,
permutation_Nums,
status='{} EARShot-Mesgarani RSA based on {}'.format(
self.data_Type, metric_Type))
print()
if self.data_Type == 'PSI':
phoneme_Feature_Distance_Dict = {
metric_Type: pairwise_distances(
self.phoneme_Feature_Array.astype(np.float64),
metric=metric_Type)
for metric_Type in self.mertic_Type_List
}
rsa_Dict['EARShot', 'Phoneme_Feature'] = {
metric_Type:
self.RSA_Calc(self.data_Array,
phoneme_Feature_Distance_Dict[metric_Type],
metric_Type, False)
for metric_Type in self.mertic_Type_List
}
permutation_Cor_List_Dict['EARShot', 'Phoneme_Feature'] = {}
for metric_Type in self.mertic_Type_List:
permutation_Cor_List_Dict['EARShot',
'Phoneme_Feature'][metric_Type] = []
for index in range(permutation_Nums):
permutation_Cor_List_Dict[
'EARShot', 'Phoneme_Feature'][metric_Type].append(
self.RSA_Calc(
self.data_Array,
phoneme_Feature_Distance_Dict[metric_Type],
metric_Type, True)[1])
progress(index + 1,
permutation_Nums,
status='{} EARShot-P&F RSA based on {}'.format(
self.data_Type, metric_Type))
print()
rsa_Dict['Phoneme_Feature', 'Mesgarani'] = {
metric_Type:
self.RSA_Calc(self.phoneme_Feature_Array,
self.mestarani_Distance_Dict[metric_Type],
metric_Type, False)
for metric_Type in self.mertic_Type_List
}
permutation_Cor_List_Dict['Phoneme_Feature', 'Mesgarani'] = {}
for metric_Type in self.mertic_Type_List:
permutation_Cor_List_Dict['Phoneme_Feature',
'Mesgarani'][metric_Type] = []
for index in range(permutation_Nums):
permutation_Cor_List_Dict[
'Phoneme_Feature', 'Mesgarani'][metric_Type].append(
self.RSA_Calc(
self.phoneme_Feature_Array,
self.mestarani_Distance_Dict[metric_Type],
metric_Type, True)[1])
progress(index + 1,
permutation_Nums,
status='{} P&F-Mesgarani RSA based on {}'.format(
self.data_Type, metric_Type))
print()
for data_Label, base_Label in [('EARShot', 'Mesgarani')] + ([
('EARShot', 'Phoneme_Feature'), ('Phoneme_Feature', 'Mesgarani')
] if self.data_Type == 'PSI' else []):
for metric_Type in self.mertic_Type_List:
p_Value = (1 - len(
np.less(
permutation_Cor_List_Dict[data_Label,
base_Label][metric_Type],
rsa_Dict[data_Label, base_Label][metric_Type][1])) /
len(permutation_Cor_List_Dict[
data_Label, base_Label][metric_Type]))
fig = self.Plot_RDM(
dm=rsa_Dict[data_Label, base_Label][metric_Type][0],
label_List=sort_List_Dict[self.data_Type],
metric=metric_Type,
fig_title=
'{0} {1} DSM: \n {2} cor: {3:.03f} \n Permutation cor: {4:.03f} \n Permutation test: p = {5:.03f}'
.format(
data_Label, self.data_Type, base_Label,
rsa_Dict[data_Label, base_Label][metric_Type][1],
np.mean(permutation_Cor_List_Dict[
data_Label, base_Label][metric_Type]), p_Value))
fig.savefig(os.path.join(
self.export_Path,
'RSA.{}_to_{}.{}.{}.png'.format(data_Label, base_Label,
self.data_Type,
metric_Type)),
dpi=300)
plt.close()
extract_List = [','.join(self.mertic_Type_List)]
extract_List.append(','.join([
'{}'.format(rsa_Dict[data_Label, base_Label][metric_Type][1])
for metric_Type in self.mertic_Type_List
]))
with open(
os.path.join(
self.export_Path, 'RSA.{}_to_{}.{}.Actual.csv'.format(
data_Label, base_Label, self.data_Type)),
'w') as f:
f.write('\n'.join(extract_List))
extract_List = [','.join(self.mertic_Type_List)]
extract_List.extend([
','.join(['{}'.format(x) for x in permutation_List])
for permutation_List in zip(*[
permutation_Cor_List_Dict[data_Label,
base_Label][metric_Type]
for metric_Type in self.mertic_Type_List
])
])
with open(
os.path.join(
self.export_Path, 'RSA.{}_to_{}.{}.Shuffle.csv'.format(
data_Label, base_Label, self.data_Type)),
'w') as f:
f.write('\n'.join(extract_List))
def Analysis(self, batch_Steps=200):
result_File_List = sorted([ #Result files sorting
os.path.join(self.result_Path, 'Test', x).replace('\\', '/')
for x in os.listdir(
os.path.join(self.result_Path, 'Test').replace('\\', '/'))
if x.endswith('.pickle') and x != 'Metadata.pickle'
])
reaction_Times = [
'\t'.join([
'{}'.format(x) for x in [
'Epoch', 'Word', 'Identifier', 'Pattern_Type',
'Pronunciation', 'Pronunciation_Length',
'Uniqueness_Point', 'Cohort_N', 'Rhyme_N',
'Neighborhood_N', 'Onset_Absolute_RT', 'Onset_Relative_RT',
'Onset_Time_Dependent_RT', 'Offset_Absolute_RT',
'Offset_Relative_RT', 'Offset_Time_Dependent_RT'
]
])
]
category_Flows = [
'\t'.join([
'{}'.format(x) for x in [
'Epoch', 'Word', 'Identifier', 'Pattern_Type',
'Pronunciation', 'Pronunciation_Length',
'Uniqueness_Point', 'Cohort_N', 'Rhyme_N',
'Neighborhood_N', 'Category', 'Category_Count', 'Accuracy'
] + list(range(self.max_Step))
])
]
for result_File in result_File_List:
with open(result_File, 'rb') as f:
result_Dict = pickle.load(f)
epoch = result_Dict['Epoch']
infos = result_Dict['Info']
outputs = result_Dict['Result'] #[Batch, Steps, Dims]
for index, (output,
(word, identifier,
pattern_Type)) in enumerate(zip(outputs, infos)):
data = self.Data_Generate(output, word, identifier,
batch_Steps) #[Num_Words, Steps]
rt_Dict = self.RT_Generate(word, identifier, data)
category_Flow_Dict = self.Category_Flow_Generate(word, data)
reaction_Times.append('\t'.join([
'{}'.format(x) for x in [
epoch, word, identifier, pattern_Type, '.'.join(
self.pattern_Metadata_Dict['Pronunciation_Dict']
[word]),
len(self.pattern_Metadata_Dict['Pronunciation_Dict']
[word]), self.adjusted_Length_Dict[word],
len(self.category_Dict[word, 'Cohort']),
len(self.category_Dict[word, 'Rhyme']),
len(self.category_Dict[word, 'DAS_Neighborhood']),
rt_Dict['Onset', 'Absolute'], rt_Dict['Onset',
'Relative'],
rt_Dict['Onset',
'Time_Dependent'], rt_Dict['Offset',
'Absolute'],
rt_Dict['Offset',
'Relative'], rt_Dict['Offset',
'Time_Dependent']
]
]))
for category in [
"Target", "Cohort", "Rhyme", "Unrelated", "Other_Max"
]:
if category == "Other_Max":
category_Count = np.nan
else:
category_Count = len(self.category_Dict[word,
category])
category_Flows.append('\t'.join([
'{}'.format(x) for x in [
epoch, word, identifier, pattern_Type, '.'.join(
self.
pattern_Metadata_Dict['Pronunciation_Dict']
[word]),
len(self.
pattern_Metadata_Dict['Pronunciation_Dict']
[word]), self.adjusted_Length_Dict[word],
len(self.category_Dict[word, 'Cohort']),
len(self.category_Dict[word, 'Rhyme']),
len(self.category_Dict[word, 'DAS_Neighborhood']),
category, category_Count,
not np.isnan(rt_Dict["Onset", "Time_Dependent"])
] + [
'{:.5f}'.format(x)
for x in category_Flow_Dict[category]
]
]))
progress(index + 1, outputs.shape[0], status=result_File)
print()
with open(
os.path.join(self.result_Path, 'Test',
'RTs.txt').replace('\\', '/'), 'w') as f:
f.write('\n'.join(reaction_Times))
with open(
os.path.join(self.result_Path, 'Test',
'Category_Flows.txt').replace('\\', '/'),
'w') as f:
f.write('\n'.join(category_Flows))
for j in range(n):
X[:, j] = (X[:, j] - X[:, j].mean())
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.2,
random_state=11)
all_theta = np.zeros((k, n + 1))
#Entrenamiento
#OneVsAll
print("Entrenando")
i = 0
for flor in Species:
progress(i + 1, k)
tmp_y = np.array(y_train == flor, dtype=int)
optTheta = logisticRegression(X_train, tmp_y, np.zeros((n + 1, 1))).x
all_theta[i] = optTheta
i += 1
#Predicciones
P = sigmoid(X_test.dot(all_theta.T))
p = [Species[np.argmax(P[i, :])] for i in range(X_test.shape[0])]
s = sum(np.array(p == y_test, dtype=int))
print("\n\n")
print("Test Accuracy ", (s / X_test.shape[0]) * 100, "%")
