custom_objects.py

from tensorflow.keras.models import load_model
import tensorflow as tf
from tensorflow.keras.saving import register_keras_serializable
from tensorflow.keras import layers, models, backend as K

@register_keras_serializable()
class SSTAmplifier(tf.keras.layers.Layer):
    def __init__(self, threshold=28.0, scale=0.1, **kwargs):
        super().__init__(**kwargs)
        self.threshold = threshold
        self.scale = scale

    def call(self, inputs):
        sst = inputs[:, 0]
        factor = tf.sigmoid((sst - self.threshold) * self.scale)
        mod = 1.0 + 0.3 * factor
        return tf.expand_dims(mod, -1)

@register_keras_serializable()
class ShearSuppressor(tf.keras.layers.Layer):
    def __init__(self, threshold=14.0, scale=0.2, **kwargs):
        super().__init__(**kwargs)
        self.threshold = threshold
        self.scale = scale

    def call(self, inputs):
        shear = inputs[:, 3]
        suppress = tf.sigmoid((self.threshold - shear) * self.scale)
        mod = 1.0 - 0.25 * suppress
        return tf.expand_dims(mod, -1)

@register_keras_serializable()
class VorticityActivator(tf.keras.layers.Layer):
    def __init__(self, threshold=1.2, scale=1.0, **kwargs):
        super().__init__(**kwargs)
        self.threshold = threshold
        self.scale = scale

    def call(self, inputs):
        vort = inputs[:, 4]
        activate = tf.sigmoid((vort - self.threshold) * self.scale)
        mod = 1.0 + 0.2 * activate
        return tf.expand_dims(mod, -1)

@register_keras_serializable()
class ModulationMixer(tf.keras.layers.Layer):
    def call(self, inputs):
        sst_mod, shear_mod, vort_mod = inputs
        product = sst_mod * shear_mod * vort_mod
        smooth = 1.0 + 0.25 * tf.tanh(product - 1.0)
        return smooth

CUSTOM_OBJECTS = {
    'ModulationMixer': ModulationMixer,
    'VorticityActivator': VorticityActivator,
    'ShearSuppressor': ShearSuppressor,
    'SSTAmplifier': SSTAmplifier
}