Running LinearClassifier.fit with SparseTensor - tensorflow

Question

I'm trying to create a LinearClassifer with a sparse binary numpy coo matrix (reports) using a SparseTensor. This is with TensorFlow 0.9.0
I do this as follows:
reports_indices = list()
rows,cols = reports.nonzero()
for row,col in zip(rows,cols):
reports_indices.append([row,col])
x_sparsetensor = tf.SparseTensor(
indices=reports_indices,
values=[1] * len(reports_indices),
shape=[reports.shape[0],reports.shape[1]])
The dimensions of reports is 10K by 1.5K.
I then setup the LinearClassifier as follows:
m = tf.contrib.learn.LinearClassifier()
m.fit(x=x_sparsetensor,y=response_vector.todense(),input_fn=None)
Response vector is binary and has a length of 10K. This results in the following error:
Traceback (most recent call last):
File "ddi_prr.py", line 38, in <module>
m.fit(x=x_sparsetensor,y=response_vector.todense(),input_fn=None)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py", line 173, in fit
input_fn, feed_fn = _get_input_fn(x, y, batch_size)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/contrib/learn/python/learn/estimators/estimator.py", line 67, in _get_input_fn
x, y, n_classes=None, batch_size=batch_size)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/contrib/learn/python/learn/io/data_feeder.py", line 117, in setup_train_data_feeder
X, y, n_classes, batch_size, shuffle=shuffle, epochs=epochs)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/contrib/learn/python/learn/io/data_feeder.py", line 240, in __init__
batch_size)
File "/usr/local/lib/python2.7/dist-packages/tensorflow/contrib/learn/python/learn/io/data_feeder.py", line 44, in _get_in_out_shape
x_shape = list(x_shape[1:]) if len(x_shape) > 1 else [1]
TypeError: object of type 'Tensor' has no len()
Is my construction incorrect for some reason? It seems that LinearClassifier.fit can't be instantiated with a SparseTensor for x, is that true? Thanks in advance for any help.

Answer 1

As far as I know, passing SparseTensors as x or y arguments to .fit is not supported:
x: matrix or tensor of shape [n_samples, n_features...]. Can be
iterator that returns arrays of features. The training input samples
for fitting the model. If set, input_fn must be None.
Also, SparseTensor is a sparse equivalent of Tensor -- an object representing symbolic computation to be executed. I believe what you would like to use as x is SparseTensorValue.
You can try pass it using other way of passing data to Estimator: input_fn function:
def get_input_fn(sparse_x, y):
def input_fn():
return sparse_x, y
m.fit(input_fn=get_input_fn(x, y))
if it won't work, you may try to produce the SparseTensors inside the input_fn function.