edunuke

from keras import layers
from keras import models
import tensorflow as tf


#
# generator input params
#

rand_dim = (1, 1, 2048)  # dimension of the generator's input tensor (gaussian noise)

edunuke

import keras

user_input = keras.layers.Input(shape=(100,), name='Domain_1')
user_vec = keras.layers.Flatten()(keras.layers.Embedding(11, 9, name='Domain_1_embed')(user_input))
user_vec = keras.layers.Dropout(0.5)(user_vec)

#Domain embedding
products_input = keras.layers.Input(shape=(100,), name='Domain_3')
products_vec = keras.layers.Flatten()(keras.layers.Embedding(50, 20, name='Domain_3_embed')(products_input))
products_vec = keras.layers.Dropout(0.5)(products_vec)

edunuke

import keras

#Domain embedding
user_input = keras.layers.Input(shape=(100,), name='Domain_1')
user_vec = keras.layers.Flatten()(keras.layers.Embedding(11, 9, name='Domain_1_embed')(user_input))
user_vec = keras.layers.Dropout(0.5)(user_vec)

#Domain embedding
products_input = keras.layers.Input(shape=(100,), name='Domain_3')
products_vec = keras.layers.Flatten()(keras.layers.Embedding(50, 20, name='Domain_3_embed')(products_input))

edunuke

import keras

#Domain embedding
user_input = keras.layers.Input(shape=(10,), name='Domain_1')
user_vec= keras.layers.Dropout(0.8)(keras.layers.Dense(1200, activation='relu',)(user_input))


#Domain embedding
products_input = keras.layers.Input(shape=(10,), name='Domain_2')
products_vec= keras.layers.Dropout(0.8)(keras.layers.Dense(1200, activation='relu',)(products_input))

edunuke

import keras

#Domain embedding
user_input = keras.layers.Input(shape=(10,), name='Domain_1')
user_vec = keras.layers.Flatten()(keras.layers.Embedding(11, 9, name='Domain_1_embed')(user_input))
user_vec = keras.layers.Dropout(0.5)(user_vec)

#Domain embedding
products_input = keras.layers.Input(shape=(10,), name='Domain_2')
products_vec = keras.layers.Flatten()(keras.layers.Embedding(50, 20, name='Domain_2_embed')(products_input))

edunuke

#fixed model weights (set them as non trainable)
for m in model_1.layers:
    if m.name == 'layer_13':
        m.trainable = False
    else:
        m.trainable = False
        
#add the new model to the old model
mixed = keras.layers.add([model_1.get_layer(name='layer_13').output, nn2], name='concat')
result_mix_1 = keras.layers.Dense(1, activation='relu', name='result_mixed')(mixed)

edunuke

input_shape = (500,)

#old model
input_1 = keras.layers.Input(shape=input_shape, name='input_1')
nn = keras.layers.Dense(input_shape[0]*3, activation='relu',name='layer_11')(input_1)
nn = keras.layers.Dropout(0.75)(nn)
nn = keras.layers.normalization.BatchNormalization()(nn)
nn = keras.layers.Dense(input_shape[0]*2, activation='relu',name='layer_12')(nn)
nn = keras.layers.Dropout(0.75)(nn)
nn = keras.layers.normalization.BatchNormalization()(nn)

edunuke

n_epoch = 1000 #example only
n_batch = 3000 # example only
X=[] #set up your feature space dimensions
y=[] # set your target dimesion
#method 1 for alternating training of the the added transfered layer weight
for i in range(n_epoch):
	model_mix_2.fit(X, y, epochs=1, batch_size=n_batch, verbose=1, shuffle=False)
    if not self.counter % 2:
        model_mix_2.get_layer(name='layer_11').trainable = True
    else:

edunuke

n_epoch = 1000 #example only
n_batch = 3000 #example only
X=[] # set to appropiate feature space dimmensions
y=[] # set the target
    
#method 2 for alternating training: use callback
class alternate_trainable(keras.callbacks.Callback):
    def __init__(self):
        self.counter = 0
        

edunuke

import pandas as pd
import seaborn as sns
import keras
import shap

#let's load the diamonds dataset
df=sns.load_dataset(name='diamonds')
print(df.head())
print(df.describe())