Spaces:
Running
Running
P2SAMAPA commited on
Update models.py
Browse files
models.py
CHANGED
|
@@ -10,6 +10,8 @@ Key components:
|
|
| 10 |
- Sparse categorical cross-entropy loss
|
| 11 |
"""
|
| 12 |
|
|
|
|
|
|
|
| 13 |
import numpy as np
|
| 14 |
import tensorflow as tf
|
| 15 |
from tensorflow.keras.models import Model
|
|
@@ -19,6 +21,13 @@ from tensorflow.keras.layers import (
|
|
| 19 |
Multiply, Add, Activation, Conv1D
|
| 20 |
)
|
| 21 |
from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
from tensorflow.keras.regularizers import l2
|
| 23 |
|
| 24 |
|
|
@@ -120,6 +129,11 @@ def train_tft(X_train, y_train, X_val, y_val, epochs=200):
|
|
| 120 |
Train the TFT classifier.
|
| 121 |
y_train/y_val: integer class labels (argmax of 5-day fwd returns).
|
| 122 |
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 123 |
seq_len = X_train.shape[1]
|
| 124 |
num_features = X_train.shape[2]
|
| 125 |
num_classes = len(np.unique(y_train))
|
|
|
|
| 10 |
- Sparse categorical cross-entropy loss
|
| 11 |
"""
|
| 12 |
|
| 13 |
+
import random
|
| 14 |
+
import os
|
| 15 |
import numpy as np
|
| 16 |
import tensorflow as tf
|
| 17 |
from tensorflow.keras.models import Model
|
|
|
|
| 21 |
Multiply, Add, Activation, Conv1D
|
| 22 |
)
|
| 23 |
from tensorflow.keras.callbacks import EarlyStopping, ReduceLROnPlateau
|
| 24 |
+
|
| 25 |
+
# ββ Fixed random seed β ensures reproducible results across runs βββββββββββββ
|
| 26 |
+
SEED = 42
|
| 27 |
+
random.seed(SEED)
|
| 28 |
+
np.random.seed(SEED)
|
| 29 |
+
tf.random.set_seed(SEED)
|
| 30 |
+
os.environ["PYTHONHASHSEED"] = str(SEED)
|
| 31 |
from tensorflow.keras.regularizers import l2
|
| 32 |
|
| 33 |
|
|
|
|
| 129 |
Train the TFT classifier.
|
| 130 |
y_train/y_val: integer class labels (argmax of 5-day fwd returns).
|
| 131 |
"""
|
| 132 |
+
# Re-apply seed immediately before model build for full reproducibility
|
| 133 |
+
random.seed(SEED)
|
| 134 |
+
np.random.seed(SEED)
|
| 135 |
+
tf.random.set_seed(SEED)
|
| 136 |
+
|
| 137 |
seq_len = X_train.shape[1]
|
| 138 |
num_features = X_train.shape[2]
|
| 139 |
num_classes = len(np.unique(y_train))
|