def train_without_pretrain():
neumf_model = NeuMF(n_user,
n_item,
gmf_dim=8,
mlp_dim=32,
layers=[32, 16, 8],
l2=0)
train(neumf_model, train_data, test_data, topk_data, epochs=50, batch=512)
def train_without_pretrain(n_user, n_item, train_data, test_data, topk_data,
gmf_dim, mlp_dim, layers, l2):
neumf_model, _, _ = NeuMF_model(n_user,
n_item,
gmf_dim=gmf_dim,
mlp_dim=mlp_dim,
layers=layers,
l2=l2)
train(neumf_model, train_data, test_data, topk_data, epochs=10, batch=512)
def train_with_pretrain(n_user, n_item, train_data, test_data, topk_data,
gmf_dim, mlp_dim, layers, l2):
neumf_model, gmf_model, mlp_model = NeuMF_model(n_user,
n_item,
gmf_dim=gmf_dim,
mlp_dim=mlp_dim,
layers=layers,
l2=l2)
print('预训练GMF部分')
train(gmf_model, train_data, test_data, topk_data, epochs=10, batch=512)
print('预训练MLP部分')
train(mlp_model, train_data, test_data, topk_data, epochs=10, batch=512)
out_kernel = tf.concat((gmf_model.get_layer('gmf_out').get_weights()[0],
mlp_model.get_layer('mlp_out').get_weights()[0]),
0)
out_bias = gmf_model.get_layer('gmf_out').get_weights(
)[1] + mlp_model.get_layer('mlp_out').get_weights()[1]
neumf_model.get_layer('out').set_weights(
[out_kernel * 0.5, out_bias * 0.5])
test(neumf_model, train_data, test_data, topk_data, batch=512)
train(neumf_model,
train_data,
test_data,
topk_data,
optimizer=tf.keras.optimizers.SGD(0.0001),
epochs=10,
batch=512)
def train_with_pretrain():
gmf_model = GMF(n_user, n_item, dim=8, l2=0)
train(gmf_model, train_data, test_data, topk_data, epochs=10, batch=512)
mlp_model = MLP(n_user, n_item, dim=32, layers=[32, 16, 8], l2=0)
train(mlp_model, train_data, test_data, topk_data, epochs=10, batch=512)
neumf_model = NeuMF(n_user,
n_item,
gmf_dim=8,
mlp_dim=32,
layers=[32, 16, 8],
l2=0)
neumf_model(tf.constant([[0, 0]]))
combine_weights(neumf_model, gmf_model, mlp_model)
train(neumf_model,
train_data,
test_data,
topk_data,
optimizer_class=None,
epochs=1,
batch=512)
train(neumf_model, train_data, test_data, topk_data, epochs=30, batch=512)
if __name__ == '__main__':
import Recommender_System.utility.gpu_memory_growth
from Recommender_System.data import data_loader, data_process
from Recommender_System.algorithm.MLP.model import MLP_model
from Recommender_System.algorithm.train import train
n_user, n_item, train_data, test_data, topk_data = data_process.pack(
data_loader.ml100k)
model = MLP_model(n_user,
n_item,
dim=32,
layers=[64, 64, 64],
l2=0,
dropout=0.3)
train(model, train_data, test_data, topk_data, epochs=30, batch=512)
if __name__ == '__main__':
import Recommender_System.utility.gpu_memory_growth
from Recommender_System.data import data_loader, data_process
from Recommender_System.algorithm.DeepFM.model import DeepFM_model
from Recommender_System.algorithm.train import train
n_user, n_item, train_data, test_data, topk_data = data_process.pack(data_loader.ml100k)
model = DeepFM_model(n_user, n_item, dim=8, layers=[16, 16, 16], l2=1e-5)
train(model, train_data, test_data, topk_data, epochs=10)
import tensorflow as tf
from Recommender_System.data import data_loader, data_process
from Recommender_System.algorithm.FM.model import FM_model
from Recommender_System.algorithm.GMF.model import GMF_model
from Recommender_System.algorithm.LFM.model import LFM_model
from Recommender_System.algorithm.MLP.model import MLP_model
from Recommender_System.algorithm.NeuMF.model import NeuMF_model
from Recommender_System.algorithm.DeepFM.model import DeepFM_model
from Recommender_System.algorithm.train import train
n_user, n_item, train_data, test_data, topk_data = data_process.pack(data_loader.ml100k)
dim = 16
model = FM_model(n_user, n_item, dim=dim, l2=0)
train(model, train_data, test_data, topk_data, epochs=10)
model = GMF_model(n_user, n_item, dim=dim, l2=0)
train(model, train_data, test_data, topk_data, epochs=10)
model = LFM_model(n_user, n_item, dim=dim, l2=0)
train(model, train_data, test_data, topk_data, loss_object=tf.losses.MeanSquaredError(), epochs=10)
model = MLP_model(n_user, n_item, dim=dim * 2, layers=[dim * 2, dim, dim // 2], l2=0)
train(model, train_data, test_data, topk_data, epochs=10)
model, _, _ = NeuMF_model(n_user, n_item, gmf_dim=dim // 2, mlp_dim=dim * 2, layers=[dim * 2, dim, dim // 2], l2=0)
train(model, train_data, test_data, topk_data, epochs=10)
model = DeepFM_model(n_user, n_item, dim // 2, layers=[dim, dim, dim], l2=0)
train(model, train_data, test_data, topk_data, epochs=10)