基于 Tensorflow 和深度神经网络，对 cenim 项目中搜集到的用户对电影的评价进行分析.zip

# -*- coding: utf-8 -*- # Module: cenim.tests.dnn.train import os.path import numpy as np import tensorflow as tf from cenim.data import input_data, MOVIE_FEATURE_SIZE from cenim.models import DnnClassifier from cenim.utils import save_data # Constants MODEL_PATH = './data/models/dnn.ckpt' # Global Variables __saver = None __weights = None __biases = None def build_model(sess): global __weights, __biases print('Building DNN model...') n_input = MOVIE_FEATURE_SIZE * 4 n_classes = 3 # Build a NN classifier dnn = DnnClassifier(sess, n_input, n_classes) (weights, biases) = dnn.add_output_layer(tf.nn.softmax) __weights = weights __biases = biases print('[DONE] - DNN model has been built.\n') return dnn def load_model(sess): print("Loading model from file: %s..." % MODEL_PATH) __saver.restore(sess, MODEL_PATH) pass def save_model(sess): __saver.save(sess, MODEL_PATH) print("[Done] Model saved in file: %s." % MODEL_PATH) def train(sess, dnn): # Generate datasets datasets = input_data.gen_datasets(0.8) # Start Training N_ITERATION = 10000 print('Training(' + str(N_ITERATION) + ' steps)...') for i in range(N_ITERATION): (X, Y) = datasets.train.next_batch(100) dnn.train(X, Y) if i % (N_ITERATION / 5) == 0: test_x, test_y = datasets.test accuracy = dnn.compute_accuracy(test_x, test_y) print(str(i * 100 / N_ITERATION) + '% completed with accuracy ' + str(accuracy * 100) + '%') print('[Done] 100% completed') save_model(sess) def run(): global __saver with tf.Session() as sess: # Build DNN and initialize Tensorflow dnn = build_model(sess) # Restore model from local or initialize for the # first time __saver = tf.train.Saver() if not(os.path.exists(MODEL_PATH + '.index')): sess.run(tf.global_variables_initializer()) else: load_model(sess) # Start training train(sess, dnn) # Save model (weights, biases) = sess.run([__weights, __biases]) save_data({ 'weights': weights.tolist(), 'biases': biases.tolist() }, 'models/checkpoint') save_model(sess)