Question

在一个简单的例子中，我尝试使用TensorFlow LinearRegressor，但结果不正确。有什么建议吗？

import tensorflow as tf
import numpy as np
x_data =np.array([  44.57,  42.71, 119.25, 40.83, 46.87, 71.44, 
113.5, 39.83, 39.48,
77.3,    53.32 ,  21.68 , 113.55  , 40.1  ,  77.39 , 46.01 , 
35.42 ,  93.81,
84.71,   51.7 ,   73.57,  102.21 ,  98.05 ,  99.53 ,  
98.65 ,  50.1,   108.4,
62.06,   48.34  , 71.45  , 53.21 ,  72.57 , 48.14 ,  
71.32 ,  41.01 ,  96.71,
112.09,   54.87 ,  63.17 , 44.95])
y_data= np.array([ 127.42 , 121.09 , 294.53,   96.73,  125.04, 195.08,   
287.84,  106.97,  107.94,
204.45,  116.09 ,  57.64 , 296.82 , 123.5 ,  180.11 , 116.81 ,  
96.73  ,233.71,
237.07,  130. ,   182.61 , 260.22,  238.86 , 238.02 , 
248.05, 101.41,  269.69,
156.43 , 121.27 , 172.64 , 139.62 , 203.87 , 134.78 , 
176.24 , 106.22 , 252.93,
282.96 ,141.95 ,161.  ,  123.42])

features = [tf.contrib.layers.real_valued_column("x", dimension=1)]
estimator = tf.contrib.learn.LinearRegressor(feature_columns=features,
     model_dir='./linear_estimator')
input_fn = tf.contrib.learn.io.numpy_input_fn({"x":x_data}, y_data,
     num_epochs=1000)
estimator.fit(input_fn=input_fn, steps=2000)
np.asarray([i for i in estimator.predict(x={'x': x_data})])

结果是

array（[1539.31665039,1476.55419922,4059.26489258,1413.11694336， 1616.92626953,2445.9987793,3865.2409668,1379.37365723， 1367.56335449,2643.734375,1834.57043457,766.93310547， 3866.92822266,1388.48425293,2646.77124023,1587.90698242， 1230.56567383,320083569696,2893.77197266,1779.90625， 2517.87182617,3484.27929688,3343.9074707,3393。84741211， 3364.15332031,1725.91699219,3693.15039062,2129.48681641， 1666.52893066,2446.3359375,1830.85864258,2484.12866211， 1659.78015137,2441.94946289,1419.19055176,3298.69140625， 3817.6628418,18868727652,21,666,94165039,1552.13916016]，dtype = float32）

Answer 1

现在可行。

我删除了导出目录并拥有新版本1.1.0。

感谢所有答案。

为什么TensorFlow LinearRegressor预测数字过大？

1 个答案: