Spaces:
Sleeping
Sleeping
xgboost
Browse files- README.md +0 -1
- app.py +114 -210
- best_v2f_generalist.keras → ml_XGBoost_v1_immed.joblib +2 -2
- requirements.txt +4 -2
README.md
CHANGED
|
@@ -7,7 +7,6 @@ sdk: gradio
|
|
| 7 |
sdk_version: 6.13.0
|
| 8 |
app_file: app.py
|
| 9 |
pinned: false
|
| 10 |
-
python_version: "3.11"
|
| 11 |
---
|
| 12 |
|
| 13 |
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
|
| 7 |
sdk_version: 6.13.0
|
| 8 |
app_file: app.py
|
| 9 |
pinned: false
|
|
|
|
| 10 |
---
|
| 11 |
|
| 12 |
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
app.py
CHANGED
|
@@ -1,23 +1,21 @@
|
|
| 1 |
import os
|
| 2 |
-
|
| 3 |
-
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
|
| 4 |
-
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2"
|
| 5 |
-
|
| 6 |
import numpy as np
|
| 7 |
import librosa
|
| 8 |
-
import
|
| 9 |
-
from scipy.stats import kurtosis
|
| 10 |
import gradio as gr
|
| 11 |
-
import
|
| 12 |
-
import zipfile
|
| 13 |
-
import tempfile
|
| 14 |
import warnings
|
| 15 |
|
| 16 |
-
warnings.filterwarnings('ignore'
|
| 17 |
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
LABEL_MAP_INVERSE = {
|
| 23 |
0: 'Machine 1_Normal', 1: 'Machine 1_Abnormal',
|
|
@@ -27,205 +25,103 @@ LABEL_MAP_INVERSE = {
|
|
| 27 |
|
| 28 |
|
| 29 |
# ==========================================
|
| 30 |
-
#
|
| 31 |
# ==========================================
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
self.sr = target_sr
|
| 35 |
-
self.n_fft = n_fft
|
| 36 |
-
self.hop_length = hop_length
|
| 37 |
-
self.n_mels = n_mels
|
| 38 |
-
self.n_mfcc = n_mfcc
|
| 39 |
-
|
| 40 |
-
def _apply_highpass_filter(self, y, cutoff=60.0):
|
| 41 |
-
nyquist = 0.5 * self.sr
|
| 42 |
-
normal_cutoff = cutoff / nyquist
|
| 43 |
-
if normal_cutoff >= 1.0:
|
| 44 |
-
return y
|
| 45 |
-
b, a = scipy.signal.butter(4, normal_cutoff, btype='high', analog=False)
|
| 46 |
-
return scipy.signal.filtfilt(b, a, y)
|
| 47 |
-
|
| 48 |
-
def _truncate_silence(self, y, top_db=25):
|
| 49 |
-
y_trimmed, _ = librosa.effects.trim(y, top_db=top_db, frame_length=self.n_fft, hop_length=self.hop_length)
|
| 50 |
-
return y_trimmed
|
| 51 |
-
|
| 52 |
-
def _mean_variance_normalize(self, y):
|
| 53 |
-
return (y - np.mean(y)) / (np.std(y) + 1e-8)
|
| 54 |
-
|
| 55 |
-
def process_audio(self, file_path):
|
| 56 |
-
y, _ = librosa.load(file_path, sr=self.sr)
|
| 57 |
-
y = self._apply_highpass_filter(y)
|
| 58 |
-
y = self._truncate_silence(y, top_db=25)
|
| 59 |
-
|
| 60 |
-
if len(y) == 0:
|
| 61 |
-
raise ValueError(f"Silence only: {file_path}")
|
| 62 |
-
|
| 63 |
-
y = self._mean_variance_normalize(y)
|
| 64 |
-
|
| 65 |
-
mel_spec = librosa.feature.melspectrogram(
|
| 66 |
-
y=y, sr=self.sr, n_fft=self.n_fft, hop_length=self.hop_length, n_mels=self.n_mels
|
| 67 |
-
)
|
| 68 |
-
log_mel_spec = librosa.power_to_db(mel_spec, ref=np.max)
|
| 69 |
-
return log_mel_spec
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
def pad_or_truncate(spectrogram, max_frames):
|
| 73 |
-
if spectrogram.shape[1] > max_frames:
|
| 74 |
-
return spectrogram[:, :max_frames]
|
| 75 |
-
elif spectrogram.shape[1] < max_frames:
|
| 76 |
-
pad_width = max_frames - spectrogram.shape[1]
|
| 77 |
-
return np.pad(spectrogram, pad_width=((0, 0), (0, pad_width)), mode='constant')
|
| 78 |
-
return spectrogram
|
| 79 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
|
| 81 |
-
#
|
| 82 |
-
|
| 83 |
-
# ==========================================
|
| 84 |
-
class NormLayer(tf.keras.layers.Layer):
|
| 85 |
-
"""Replaces Lambda(lambda t: t / 80.0) used for spectrogram normalization."""
|
| 86 |
-
def call(self, x):
|
| 87 |
-
return x / 80.0
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
class FreqReduceLayer(tf.keras.layers.Layer):
|
| 91 |
-
"""Replaces Lambda(lambda t: tf.reduce_mean(t, axis=1)) used to collapse frequency."""
|
| 92 |
-
def call(self, x):
|
| 93 |
-
return tf.reduce_mean(x, axis=1)
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
class SpecAugmentLayer(tf.keras.layers.Layer):
|
| 97 |
-
"""Training-only augmentation. Passes through during inference."""
|
| 98 |
-
def __init__(self, freq_mask_param=15, time_mask_param=30, **kwargs):
|
| 99 |
-
super().__init__(**kwargs)
|
| 100 |
-
self.freq_mask_param = freq_mask_param
|
| 101 |
-
self.time_mask_param = time_mask_param
|
| 102 |
-
|
| 103 |
-
def call(self, inputs, training=None):
|
| 104 |
-
if not training:
|
| 105 |
-
return inputs
|
| 106 |
-
freq_max = tf.shape(inputs)[1]
|
| 107 |
-
time_max = tf.shape(inputs)[2]
|
| 108 |
-
f = tf.random.uniform([], minval=0, maxval=self.freq_mask_param, dtype=tf.int32)
|
| 109 |
-
f0 = tf.random.uniform([], minval=0, maxval=freq_max - f, dtype=tf.int32)
|
| 110 |
-
freq_indices = tf.range(freq_max)
|
| 111 |
-
freq_mask = tf.logical_or(freq_indices < f0, freq_indices >= f0 + f)
|
| 112 |
-
freq_mask = tf.cast(freq_mask, inputs.dtype)
|
| 113 |
-
freq_mask = tf.reshape(freq_mask, [1, -1, 1, 1])
|
| 114 |
-
inputs = inputs * freq_mask
|
| 115 |
-
t = tf.random.uniform([], minval=0, maxval=self.time_mask_param, dtype=tf.int32)
|
| 116 |
-
t0 = tf.random.uniform([], minval=0, maxval=time_max - t, dtype=tf.int32)
|
| 117 |
-
time_indices = tf.range(time_max)
|
| 118 |
-
time_mask = tf.logical_or(time_indices < t0, time_indices >= t0 + t)
|
| 119 |
-
time_mask = tf.cast(time_mask, inputs.dtype)
|
| 120 |
-
time_mask = tf.reshape(time_mask, [1, 1, -1, 1])
|
| 121 |
-
inputs = inputs * time_mask
|
| 122 |
-
return inputs
|
| 123 |
-
|
| 124 |
-
def get_config(self):
|
| 125 |
-
config = super().get_config()
|
| 126 |
-
config.update({
|
| 127 |
-
"freq_mask_param": self.freq_mask_param,
|
| 128 |
-
"time_mask_param": self.time_mask_param,
|
| 129 |
-
})
|
| 130 |
-
return config
|
| 131 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 132 |
|
| 133 |
-
#
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
se = tf.keras.layers.Reshape([1, 1, filters])(se)
|
| 141 |
-
return x * se
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
def build_v2f():
|
| 145 |
-
inp = tf.keras.Input(shape=(N_MELS, MAX_TIME_FRAMES, 1))
|
| 146 |
-
|
| 147 |
-
x = SpecAugmentLayer(freq_mask_param=15, time_mask_param=30)(inp)
|
| 148 |
-
x = NormLayer()(x)
|
| 149 |
-
|
| 150 |
-
# Block 1
|
| 151 |
-
x = tf.keras.layers.Conv2D(32, (3, 3), padding='same',
|
| 152 |
-
kernel_regularizer=tf.keras.regularizers.l2(1e-4))(x)
|
| 153 |
-
x = tf.keras.layers.BatchNormalization()(x)
|
| 154 |
-
x = tf.keras.layers.Activation('relu')(x)
|
| 155 |
-
x = se_block(x, 32)
|
| 156 |
-
x = tf.keras.layers.MaxPooling2D(pool_size=(2, 1))(x)
|
| 157 |
-
x = tf.keras.layers.Dropout(0.2)(x)
|
| 158 |
-
|
| 159 |
-
# Block 2
|
| 160 |
-
x = tf.keras.layers.Conv2D(64, (3, 3), padding='same',
|
| 161 |
-
kernel_regularizer=tf.keras.regularizers.l2(1e-4))(x)
|
| 162 |
-
x = tf.keras.layers.BatchNormalization()(x)
|
| 163 |
-
x = tf.keras.layers.Activation('relu')(x)
|
| 164 |
-
x = se_block(x, 64)
|
| 165 |
-
x = tf.keras.layers.MaxPooling2D(pool_size=(2, 1))(x)
|
| 166 |
-
x = tf.keras.layers.Dropout(0.2)(x)
|
| 167 |
-
|
| 168 |
-
# Block 3
|
| 169 |
-
x = tf.keras.layers.Conv2D(128, (3, 3), padding='same',
|
| 170 |
-
kernel_regularizer=tf.keras.regularizers.l2(1e-4))(x)
|
| 171 |
-
x = tf.keras.layers.BatchNormalization()(x)
|
| 172 |
-
x = tf.keras.layers.Activation('relu')(x)
|
| 173 |
-
x = se_block(x, 128)
|
| 174 |
-
x = tf.keras.layers.MaxPooling2D(pool_size=(2, 1))(x)
|
| 175 |
-
x = tf.keras.layers.Dropout(0.2)(x)
|
| 176 |
-
|
| 177 |
-
# Bridge: reduce channels and collapse frequency
|
| 178 |
-
x = tf.keras.layers.Conv2D(64, (1, 1), padding='same', activation='relu')(x)
|
| 179 |
-
x = FreqReduceLayer()(x)
|
| 180 |
-
|
| 181 |
-
# BiLSTM
|
| 182 |
-
x = tf.keras.layers.Bidirectional(
|
| 183 |
-
tf.keras.layers.LSTM(64, return_sequences=True)
|
| 184 |
-
)(x)
|
| 185 |
-
x = tf.keras.layers.Dropout(0.3)(x)
|
| 186 |
-
|
| 187 |
-
# MultiHead Attention
|
| 188 |
-
x = tf.keras.layers.MaxPooling1D(pool_size=4)(x)
|
| 189 |
-
x = tf.keras.layers.MultiHeadAttention(num_heads=4, key_dim=32)(x, x)
|
| 190 |
-
x = tf.keras.layers.GlobalAveragePooling1D()(x)
|
| 191 |
-
x = tf.keras.layers.Dropout(0.4)(x)
|
| 192 |
-
|
| 193 |
-
out = tf.keras.layers.Dense(NUM_CLASSES, activation='softmax')(x)
|
| 194 |
-
return tf.keras.Model(inputs=inp, outputs=out)
|
| 195 |
|
| 196 |
|
| 197 |
# ==========================================
|
| 198 |
-
#
|
| 199 |
# ==========================================
|
| 200 |
-
def
|
| 201 |
-
|
| 202 |
-
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
| 213 |
-
|
| 214 |
-
|
| 215 |
-
|
| 216 |
-
|
| 217 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 218 |
|
| 219 |
|
| 220 |
# ==========================================
|
| 221 |
-
#
|
| 222 |
# ==========================================
|
| 223 |
-
model
|
| 224 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 225 |
print(f"Warning: Could not load model. Error: {model_error}")
|
| 226 |
|
| 227 |
-
preprocessor = MachineListenerPreprocessor()
|
| 228 |
-
|
| 229 |
|
| 230 |
# ==========================================
|
| 231 |
# PREDICTION
|
|
@@ -237,19 +133,27 @@ def predict(audio_filepath):
|
|
| 237 |
return "Please upload an audio file."
|
| 238 |
|
| 239 |
try:
|
| 240 |
-
|
| 241 |
-
|
| 242 |
-
|
| 243 |
-
#
|
| 244 |
-
|
| 245 |
-
|
| 246 |
-
|
| 247 |
-
|
| 248 |
-
|
| 249 |
-
|
| 250 |
-
|
| 251 |
-
|
| 252 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 253 |
|
| 254 |
except Exception as e:
|
| 255 |
return f"Error processing file: {str(e)}"
|
|
|
|
| 1 |
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
import numpy as np
|
| 3 |
import librosa
|
| 4 |
+
import joblib
|
|
|
|
| 5 |
import gradio as gr
|
| 6 |
+
from scipy import stats as scipy_stats
|
|
|
|
|
|
|
| 7 |
import warnings
|
| 8 |
|
| 9 |
+
warnings.filterwarnings('ignore')
|
| 10 |
|
| 11 |
+
# ==========================================
|
| 12 |
+
# CONFIGURATION (must match training notebook)
|
| 13 |
+
# ==========================================
|
| 14 |
+
TARGET_SR = 48000
|
| 15 |
+
TARGET_DURATION = 10.0
|
| 16 |
+
TARGET_LENGTH = int(TARGET_SR * TARGET_DURATION)
|
| 17 |
+
TRIM_TOP_DB = 20
|
| 18 |
+
N_MFCC = 20
|
| 19 |
|
| 20 |
LABEL_MAP_INVERSE = {
|
| 21 |
0: 'Machine 1_Normal', 1: 'Machine 1_Abnormal',
|
|
|
|
| 25 |
|
| 26 |
|
| 27 |
# ==========================================
|
| 28 |
+
# PREPROCESSING (from notebook 02)
|
| 29 |
# ==========================================
|
| 30 |
+
def preprocess_audio(file_path):
|
| 31 |
+
y, sr = librosa.load(file_path, sr=TARGET_SR)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
|
| 33 |
+
# Trim silence
|
| 34 |
+
y_trimmed, _ = librosa.effects.trim(y, top_db=TRIM_TOP_DB)
|
| 35 |
+
if len(y_trimmed) == 0:
|
| 36 |
+
y_trimmed = y
|
| 37 |
|
| 38 |
+
# DC offset removal
|
| 39 |
+
y_centered = y_trimmed - np.mean(y_trimmed)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
+
# Z-score normalization
|
| 42 |
+
std = np.std(y_centered)
|
| 43 |
+
if std > 1e-8:
|
| 44 |
+
y_norm = y_centered / std
|
| 45 |
+
else:
|
| 46 |
+
y_norm = y_centered
|
| 47 |
|
| 48 |
+
# Fixed length
|
| 49 |
+
if len(y_norm) < TARGET_LENGTH:
|
| 50 |
+
y_norm = np.pad(y_norm, (0, TARGET_LENGTH - len(y_norm)), mode='constant')
|
| 51 |
+
else:
|
| 52 |
+
y_norm = y_norm[:TARGET_LENGTH]
|
| 53 |
+
|
| 54 |
+
return y_norm
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
|
| 56 |
|
| 57 |
# ==========================================
|
| 58 |
+
# FEATURE EXTRACTION (from notebook 04)
|
| 59 |
# ==========================================
|
| 60 |
+
def compute_stats(feature_array, prefix):
|
| 61 |
+
result = {}
|
| 62 |
+
if feature_array.ndim == 1:
|
| 63 |
+
feature_array = feature_array.reshape(1, -1)
|
| 64 |
+
|
| 65 |
+
for i in range(feature_array.shape[0]):
|
| 66 |
+
row = feature_array[i]
|
| 67 |
+
suffix = f"_{i}" if feature_array.shape[0] > 1 else ""
|
| 68 |
+
result[f"{prefix}{suffix}_mean"] = np.mean(row)
|
| 69 |
+
result[f"{prefix}{suffix}_std"] = np.std(row)
|
| 70 |
+
result[f"{prefix}{suffix}_min"] = np.min(row)
|
| 71 |
+
result[f"{prefix}{suffix}_max"] = np.max(row)
|
| 72 |
+
return result
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
def extract_ml_features(y, sr=TARGET_SR, n_mfcc=N_MFCC):
|
| 76 |
+
features = {}
|
| 77 |
+
|
| 78 |
+
# 1. MFCCs (20 × 4 = 80)
|
| 79 |
+
mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=n_mfcc)
|
| 80 |
+
features.update(compute_stats(mfccs, "mfcc"))
|
| 81 |
+
|
| 82 |
+
# 2. Spectral Centroid (4)
|
| 83 |
+
features.update(compute_stats(librosa.feature.spectral_centroid(y=y, sr=sr), "spectral_centroid"))
|
| 84 |
+
|
| 85 |
+
# 3. Spectral Flatness (4)
|
| 86 |
+
features.update(compute_stats(librosa.feature.spectral_flatness(y=y), "spectral_flatness"))
|
| 87 |
+
|
| 88 |
+
# 4. Zero-Crossing Rate (4)
|
| 89 |
+
features.update(compute_stats(librosa.feature.zero_crossing_rate(y), "zcr"))
|
| 90 |
+
|
| 91 |
+
# 5. RMS Energy (4)
|
| 92 |
+
features.update(compute_stats(librosa.feature.rms(y=y), "rms"))
|
| 93 |
+
|
| 94 |
+
# 6. Spectral Rolloff at 85% (4)
|
| 95 |
+
features.update(compute_stats(librosa.feature.spectral_rolloff(y=y, sr=sr, roll_percent=0.85), "spectral_rolloff"))
|
| 96 |
+
|
| 97 |
+
# 7. Spectral Bandwidth (4)
|
| 98 |
+
features.update(compute_stats(librosa.feature.spectral_bandwidth(y=y, sr=sr), "spectral_bandwidth"))
|
| 99 |
+
|
| 100 |
+
# 8. Spectral Contrast (7 bands × 4 = 28)
|
| 101 |
+
features.update(compute_stats(librosa.feature.spectral_contrast(y=y, sr=sr, n_bands=6), "spectral_contrast"))
|
| 102 |
+
|
| 103 |
+
# 9. Chroma STFT (12 × 4 = 48)
|
| 104 |
+
features.update(compute_stats(librosa.feature.chroma_stft(y=y, sr=sr), "chroma"))
|
| 105 |
+
|
| 106 |
+
# 10. Spectral Kurtosis (1)
|
| 107 |
+
S = np.abs(librosa.stft(y))
|
| 108 |
+
features["spectral_kurtosis"] = float(scipy_stats.kurtosis(np.mean(S, axis=1)))
|
| 109 |
+
|
| 110 |
+
return features
|
| 111 |
|
| 112 |
|
| 113 |
# ==========================================
|
| 114 |
+
# LOAD MODEL
|
| 115 |
# ==========================================
|
| 116 |
+
model = None
|
| 117 |
+
model_error = ""
|
| 118 |
+
try:
|
| 119 |
+
model = joblib.load("ml_XGBoost_v1_immed.joblib")
|
| 120 |
+
print("XGBoost model loaded successfully.")
|
| 121 |
+
except Exception as e:
|
| 122 |
+
model_error = str(e)
|
| 123 |
print(f"Warning: Could not load model. Error: {model_error}")
|
| 124 |
|
|
|
|
|
|
|
| 125 |
|
| 126 |
# ==========================================
|
| 127 |
# PREDICTION
|
|
|
|
| 133 |
return "Please upload an audio file."
|
| 134 |
|
| 135 |
try:
|
| 136 |
+
# 1. Preprocess
|
| 137 |
+
y = preprocess_audio(audio_filepath)
|
| 138 |
+
|
| 139 |
+
# 2. Extract features
|
| 140 |
+
features = extract_ml_features(y)
|
| 141 |
+
|
| 142 |
+
# 3. Convert to array in the same column order the model expects
|
| 143 |
+
feature_names = sorted(features.keys())
|
| 144 |
+
X = np.array([[features[name] for name in feature_names]])
|
| 145 |
+
|
| 146 |
+
# 4. Predict
|
| 147 |
+
predicted_class = model.predict(X)[0]
|
| 148 |
+
predicted_label = LABEL_MAP_INVERSE.get(int(predicted_class), "Unknown")
|
| 149 |
+
|
| 150 |
+
# Get probabilities if available
|
| 151 |
+
if hasattr(model, 'predict_proba'):
|
| 152 |
+
proba = model.predict_proba(X)[0]
|
| 153 |
+
confidence = float(np.max(proba))
|
| 154 |
+
return f"Prediction: {predicted_label} (Confidence: {confidence:.2f})"
|
| 155 |
+
else:
|
| 156 |
+
return f"Prediction: {predicted_label}"
|
| 157 |
|
| 158 |
except Exception as e:
|
| 159 |
return f"Error processing file: {str(e)}"
|
best_v2f_generalist.keras → ml_XGBoost_v1_immed.joblib
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2e80bff346c7895461984c34edc41667489208e66fd5de49881bb136a096d664
|
| 3 |
+
size 3011600
|
requirements.txt
CHANGED
|
@@ -1,5 +1,7 @@
|
|
| 1 |
-
tensorflow-cpu==2.16.*
|
| 2 |
gradio
|
| 3 |
librosa
|
| 4 |
-
numpy
|
| 5 |
scipy
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
gradio
|
| 2 |
librosa
|
| 3 |
+
numpy
|
| 4 |
scipy
|
| 5 |
+
scikit-learn
|
| 6 |
+
xgboost
|
| 7 |
+
joblib
|