使用SkFlow TensorFlowDNNRegressor的多个目标列

Question

我刚开始使用Tensorflow/SkFlow，我正在尝试弄清楚是否可以使用多个目标列并生成多个输出预测。

我尝试了下面的代码，但这似乎不是可接受的输入：

import numpy as np
import tensorflow.contrib.learn as skflow

# Sample data (obviously actual data would contain a lot more rows)
training_data = np.asarray( [
    np.asarray( [ 215.0, 5.0], dtype=np.float64 ),
    np.asarray( [ 283.0, 2.0], dtype=np.float64 )
], dtype=np.float64 )
training_target  = np.asarray( [
    np.asarray( [ 220.0, 210.0], dtype=np.float64 ),
    np.asarray( [ 285.0, 281.0], dtype=np.float64 )
], dtype=np.float64 )

regressor = skflow.TensorFlowDNNRegressor( hidden_units=[2,4,2] )
regressor.fit( x=training_data, y=training_target, steps=2000 )

print( regressor.predict( training_set.data )[0] )

当我运行这段代码时，我得到以下错误：

File "/some/path/anaconda/anaconda/lib/python2.7/site-packages/tensorflow/python/framework/tensor_shape.py", line 741, in assert_is_compatible_with
raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (?, 1) and (?, 2) are incompatible

有没有可能用SkFlow做这样的事情呢？

Answer 1

有一段代码使用了DNNRegressor：

import numpy as np
from sklearn.cross_validation import train_test_split
from tensorflow.contrib import learn
import tensorflow as tf
import logging
#logging.getLogger().setLevel(logging.INFO)

#Some fake data

N=200
X=np.array(range(N),dtype=np.float32)/(N/10)
X=X[:,np.newaxis]

Y=np.sin(X.squeeze())+np.random.normal(0, 0.5, N)

X_train, X_test, Y_train, Y_test = train_test_split(X, Y,
                                                    train_size=0.8,
                                                    test_size=0.2)


reg=learn.DNNRegressor(hidden_units=[10,10])
reg.fit(X_train,Y_train,steps=500)

当我测试时，如果Y_train的形状是N*1，这个代码可以工作，否则，它将失败。我不知道如何解决这个问题。

但是，我使用tflearn模块编写了一个多目标回归演示，可能会对您有所帮助。

import tflearn
import tflearn.datasets.mnist as mnist

X,Y,testX,testY = mnist.load_data(one_hot=True)

input_layer = tflearn.input_data(shape=[None, 784],name='input')
dense1 = tflearn.fully_connected(input_layer,128,name='dense1')
dense2 = tflearn.fully_connected(dense1,256,name='dense2')
final  = tflearn.fully_connected(dense2,10,activation='relu')
regression = tflearn.regression(final,optimizer='adam',
                                learning_rate=0.001,
                                loss='mean_square')

model = tflearn.DNN(regression,checkpoint_path='model.tf.ckpt')

model.fit(X,Y,n_epoch=1,
          validation_set=(testX,testY),
          show_metric=True,
          snapshot_epoch=True,
          snapshot_step=500,
          run_id='tflearnDemo')

pred = model.predict(testX)

for i in range(len(testX)):
    print('the original data: ', testY[i], \
          'the predict  data: ', pred[i])
    print("[*]============================")

ZhQ