Python实现FFM【tensorflow2.0】
本帖最后由 糖逗 于 2020-12-26 20:54 编辑论文：https://www.csie.ntu.edu.tw/~cjlin/papers/ffm.pdf
参考：https://zhuanlan.zhihu.com/p/170607706
说明：待验证
import tensorflow as tf
from tensorflow.keras import layers, optimizers
from tensorflow import keras
from sklearn.model_selection import train_test_split
import numpy as np
#将数据划分为测试集和训练集
def preprocess(x, y):
x = tf.cast(x, dtype = tf.float64)
y = tf.cast(y, dtype = tf.int64)
return x, y
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
data = load_breast_cancer()
x_train, x_test, y_train, y_test = train_test_split(data.data, data.target, test_size=0.2,
random_state = 11, stratify = data.target)
print(x_train.shape, y_train.shape, x_test.shape, y_test.shape)
train_db = tf.data.Dataset.from_tensor_slices((np.array(x_train), y_train))
train_db = train_db.shuffle(123).map(preprocess).batch(20)
test_db = tf.data.Dataset.from_tensor_slices((np.array(x_test), y_test))
test_db = test_db.map(preprocess).batch(20)
sample = next(iter(train_db))
print('sample:', sample.shape, sample.shape,
tf.reduce_min(sample), tf.reduce_max(sample))
class FFM(keras.Model):
def __init__(self, field_num, feature_field_dict, dim_num, k = 2):
super(FFM, self).__init__()
self.field_num = field_num
self.k = k
self.feature_field_dict = feature_field_dict
self.dim_num = dim_num
def build(self, input_shape):
self.fc = tf.keras.layers.Dense(units = 1,
bias_regularizer = tf.keras.regularizers.l2(0.01),
kernel_regularizer = tf.keras.regularizers.l1(0.02))
self.w = self.add_weight(shape = (input_shape[-1], self.field_num, self.k),
initializer = 'glorot_uniform',
trainable = True)
super(FFM, self).build(input_shape)
def call(self, x, training):
linear = self.fc(x)
temp = tf.cast(0, tf.float32)
temp = tf.expand_dims(temp, axis = 0)
for j1 in range(self.dim_num):
for j2 in range(j1 + 1, self.dim_num):
f1 = self.feature_field_dict
f2 = self.feature_field_dict
#[, , k] * [, , k] = [, , k] ->
ww = tf.expand_dims(tf.multiply(self.w, self.w), axis = 0)
# * = ->
xx = tf.expand_dims(tf.multiply(x[:, j1], x[:, j2]),axis = 1)
# @ =
store = tf.matmul(xx, ww)
# ->
temp += tf.reduce_mean(store, keepdims = True, axis = 1)
out = layers.Add()()
return tf.sigmoid(out)
store = {}
for i in range(30):
store = int(i / 15)
model = FFM(field_num = 2, feature_field_dict = store, dim_num = 30)
model.build((None, 30))
model.summary()
def main():
store = {}
for i in range(30):
store = int(i / 15) #实际要根据数据字段含义定义，这里只是做一个随意的分组
model = FFM(field_num = 2, feature_field_dict = store, dim_num = 30)
optimizer = optimizers.Adam(lr = 1e-2)
for epoch in range(50):
for step, (x,y) in enumerate(train_db):
with tf.GradientTape() as tape:
logits = model(x,training=True)
loss = tf.reduce_mean(tf.losses.binary_crossentropy(y, logits))
loss_regularization = []
for i in model.trainable_variables:
loss_regularization.append(tf.nn.l2_loss(i))
loss_regularization = tf.reduce_sum(tf.stack(loss_regularization))
loss = 0.001 * loss_regularization + loss
grads = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(grads, model.trainable_variables))
print(epoch, step, 'loss:', float(loss))
total_num = 0
total_correct = 0
for x,y in test_db:
pred = model(x, training=False)
pred = tf.squeeze(pred)
pred = pred > 0.5
pred = tf.cast(pred, dtype = tf.int64)
correct = tf.cast(tf.equal(pred, y), tf.int64)
correct = tf.reduce_sum(correct)
total_num += x.shape
total_correct += int(correct)
acc = total_correct / total_num
print(epoch, 'acc:', acc)
print("-"*25)
if __name__ == '__main__':
main()
页:
[1]