TensorFlow DNNRegressor的简单使用

# coding: utf-8 # 详细可以参考 http://blog.csdn.net/u010099080/article/details/72824899 import numpy as np import pandas as pd import tensorflow as tf from sklearn.datasets import load_boston from sklearn.model_selection import train_test_split import matplotlib.pyplot as plt import seaborn as sns boston = load_boston() boston_df = pd.DataFrame(np.c_[boston.data, boston.target], columns=np.append(boston.feature_names, 'MEDV')) LABEL_COLUMN = ['MEDV'] FEATURE_COLUMNS = [f for f in boston_df if not f in LABEL_COLUMN] X_train, X_test, y_train, y_test = train_test_split(boston_df[FEATURE_COLUMNS], boston_df[LABEL_COLUMN], test_size=0.3) print('训练集：{}\n测试集：{}'.format(X_train.shape, X_test.shape)) # ## 定义 FeatureColumns feature_cols = [tf.contrib.layers.real_valued_column(k) for k in FEATURE_COLUMNS] # ## 定义 regressor config = tf.contrib.learn.RunConfig(gpu_memory_fraction=0.3, log_device_placement=False) regressor = tf.contrib.learn.DNNRegressor(feature_columns=feature_cols, hidden_units=[10, 128], model_dir='./models/dnnregressor', config=config) # ## 定义 input_fn def input_fn(df, label): feature_cols = {k: tf.constant(df[k].values) for k in FEATURE_COLUMNS} label = tf.constant(label.values) return feature_cols, label def train_input_fn(): '''训练阶段使用的 input_fn''' return input_fn(X_train, y_train) def test_input_fn(): '''测试阶段使用的 input_fn''' return input_fn(X_test, y_test) # 训练 regressor.fit(input_fn=train_input_fn, steps=5000) # 测试 ev = regressor.evaluate(input_fn=test_input_fn, steps=1) print('ev: {}'.format(ev))