'Tensor' object has no attribute 'assign_add' - object

I encountered error 'Tensor' object has no attribute 'assign_add' when I try to use the assign_add or assign_sub function.
The code is shown below:
I defined two tensor t1 and t2, with the same shape, and same data type.
>>> t1 = tf.Variable(tf.ones([2,3,4],tf.int32))
>>> t2 = tf.Variable(tf.zeros([2,3,4],tf.int32))
>>> t1
<tf.Variable 'Variable_4:0' shape=(2, 3, 4) dtype=int32_ref>
>>> t2
<tf.Variable 'Variable_5:0' shape=(2, 3, 4) dtype=int32_ref>
then I use the assign_add on t1 and t2 to create t3
>>> t3 = tf.assign_add(t1,t2)
>>> t3
<tf.Tensor 'AssignAdd_4:0' shape=(2, 3, 4) dtype=int32_ref>
then I try to create a new tensor t4 using t1[1] and t2[1], which are tensors with same shape and same data type.
>>> t1[1]
<tf.Tensor 'strided_slice_23:0' shape=(3, 4) dtype=int32>
>>> t2[1]
<tf.Tensor 'strided_slice_24:0' shape=(3, 4) dtype=int32>
>>> t4 = tf.assign_add(t1[1],t2[1])
but got error,
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/Users/admin/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/state_ops.py", line 245, in assign_add
return ref.assign_add(value)
AttributeError: 'Tensor' object has no attribute 'assign_add'
same error when using assign_sub
>>> t4 = tf.assign_sub(t1[1],t2[1])
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/Users/admin/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/state_ops.py", line 217, in assign_sub
return ref.assign_sub(value)
AttributeError: 'Tensor' object has no attribute 'assign_sub'
Any idea where is wrong?
Thanks.

The error is because t1 is a tf.Variable object , while t1[1] is a tf.Tensor.(you can see this in the outputs to your print statements.).Ditto for t2 and t[[2]]
As it happens, tf.Tensor can't be mutated(it's read only) whereas tf.Variable can be(read as well as write)
see here.
Since tf.scatter_add,does an inplace addtion, it doesn't work with t1[1] and t2[1] as inputs, while there's no such problem with t1 and t2 as inputs.

What you are trying to do here is a little bit confusing. I don't think you can update slices and create a new tensor at the same time/line.
If you want to update slices before creating t4, use tf.scatter_add() (or tf.scatter_sub() or tf.scatter_update() accordingly) as suggested here. For example:
sa = tf.scatter_add(t1, [1], t2[1:2])
Then if you want to get a new tensor t4 using new t1[1] and t2[1], you can do:
with tf.control_dependencies([sa]):
t4 = tf.add(t1[1],t2[1])

Here are some examples for using tf.scatter_add and tf.scatter_sub
>>> t1 = tf.Variable(tf.ones([2,3,4],tf.int32))
>>> t2 = tf.Variable(tf.zeros([2,3,4],tf.int32))
>>> init = tf.global_variables_initializer()
>>> sess.run(init)
>>> t1.eval()
array([[[1, 1, 1, 1],
[1, 1, 1, 1],
[1, 1, 1, 1]],
[[1, 1, 1, 1],
[1, 1, 1, 1],
[1, 1, 1, 1]]], dtype=int32)
>>> t2.eval()
array([[[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]],
[[0, 0, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]]], dtype=int32)
>>> t3 = tf.scatter_add(t1,[0],[[[2,2,2,2],[2,2,2,2],[2,2,2,2]]])
>>> sess.run(t3)
array([[[3, 3, 3, 3],
[3, 3, 3, 3],
[3, 3, 3, 3]],
[[1, 1, 1, 1],
[1, 1, 1, 1],
[1, 1, 1, 1]]], dtype=int32)
>>>t4 = tf.scatter_sub(t1,[0,0,0],[t1[1],t1[1],t1[1]])
Following is another example, which can be found at https://blog.csdn.net/efforever/article/details/77073103
Because few examples illustrating scatter_xxx can be found on the web, I paste it below for reference.
import tensorflow as tf
import numpy as np
with tf.Session() as sess1:
c = tf.Variable([[1,2,0],[2,3,4]], dtype=tf.float32, name='biases')
cc = tf.Variable([[1,2,0],[2,3,4]], dtype=tf.float32, name='biases1')
ccc = tf.Variable([0,1], dtype=tf.int32, name='biases2')
#对应label的centers-diff[0--]
centers = tf.scatter_sub(c,ccc,cc)
#centers = tf.scatter_sub(c,[0,1],cc)
#centers = tf.scatter_sub(c,[0,1],[[1,2,0],[2,3,4]])
#centers = tf.scatter_sub(c,[0,0,0],[[1,2,0],[2,3,4],[1,1,1]])
#即c[0]-[1,2,0] \ c[0]-[2,3,4]\ c[0]-[1,1,1],updates要减完:indices与updates元素个数相同
a = tf.Variable(initial_value=[[0, 0, 0, 0],[0, 0, 0, 0]])
b = tf.scatter_update(a, [0, 1], [[1, 1, 0, 0], [1, 0, 4, 0]])
#b = tf.scatter_update(a, [0, 1,0], [[1, 1, 0, 0], [1, 0, 4, 0],[1, 1, 0, 1]])
init = tf.global_variables_initializer()
sess1.run(init)
print(sess1.run(centers))
print(sess1.run(b))
[[ 0. 0. 0.]
[ 0. 0. 0.]]
[[1 1 0 0]
[1 0 4 0]]
[[-3. -4. -5.]
[ 2. 3. 4.]]
[[1 1 0 1]
[1 0 4 0]]

You can also use tf.assign() as a workaround as sliced assign was implemented for it, unlike for tf.assign_add() or tf.assign_sub(), as of TensorFlow version 1.8. Please note, you can only do one slicing operation (slice into slice is not going to work) and also this is not atomic, so if there are multiple threads reading and writing to the same variable, you don't know which operation will be the last one to write unless you explicitly code for it. tf.assign_add() and tf.assign_sub() are guaranteed to be thread safe. Still, this is better that nothing: consider this code (tested):
import tensorflow as tf
t1 = tf.Variable(tf.zeros([2,3,4],tf.int32))
t2 = tf.Variable(tf.ones([2,3,4],tf.int32))
assign_op = tf.assign( t1[ 1 ], t1[ 1 ] + t2[ 1 ] )
init_op = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run( init_op )
res = sess.run( assign_op )
print( res )
will output:
[[[0 0 0 0]
[0 0 0 0]
[0 0 0 0]]
[[1 1 1 1]
[1 1 1 1]
[1 1 1 1]]]
as desired.

Related

Multiclass vs. multilabel fitting

In scikit-learn tutorials, I found the following paragraphs in the section 'Multiclass vs. multilabel fitting'.
I couldn't understand why the following codes generate the given results.
First
from sklearn.svm import SVC
from sklearn.multiclass import OneVsRestClassifier
from sklearn.preprocessing import LabelBinarizer
X = [[1, 2], [2, 4], [4, 5], [3, 2], [3, 1]]
y = [0, 0, 1, 1, 2]
classif = OneVsRestClassifier(estimator=SVC(random_state=0))
classif.fit(X, y).predict(X)
array([0, 0, 1, 1, 2])
y = LabelBinarizer().fit_transform(y)
classif.fit(X, y).predict(X)
array([[1, 0, 0],
[1, 0, 0],
[0, 1, 0],
[0, 0, 0],
[0, 0, 0]])
Next
from sklearn.preprocessing import MultiLabelBinarizer
y = [[0, 1], [0, 2], [1, 3], [0, 2, 3], [2, 4]]
y = MultiLabelBinarizer().fit_transform(y)
classif.fit(X, y).predict(X)
array([[1, 1, 0, 0, 0],
[1, 0, 1, 0, 0],
[0, 1, 0, 1, 0],
[1, 0, 1, 0, 0],
[1, 0, 1, 0, 0]])
Label binarization in scikit-learn will transform your targets and represent them in a label indicator matrix. This label indicator matrix has the shape (n_samples, n_classes) and is composed as follows:
each row represents a sample
each column represents a class
each element is 1 if the sample is labeled with the class and 0 if not
In your first example, you have a target collection with 5 samples and 3 classes. That's why transforming y with LabelBinarizer results in a 5x3 matrix. In your case, [1, 0, 0] corresponds to class 0, [0, 1, 0] corresponds to class 1 and so forth. Notice that in each row there is only one element set to 1, since each sample can have one label only.
In your next example, you have a target collection with 5 samples and 5 classes. That's why transforming y with MultiLabelBinarizer results in a 5x5 matrix. In your case, [1, 1, 0, 0, 0] corresponds to the multilabel [0, 1], [0, 1, 0, 1, 0] corresponds to the multilabel [1, 3] and so forth. The key difference to the first example is that each row can have multiple elements set to 1, because each sample can have multiple labels/classes.
The predicted values you get follow the very same pattern. They are however not equivalent to the original values in y since your classification model has obviously predicted different values. You can check this with the inverse_transform() of the binarizers:
from sklearn.preprocessing import MultiLabelBinarizer
mlb = MultiLabelBinarizer()
y = np.array([[0, 1], [0, 2], [1, 3], [0, 2, 3], [2, 4]])
y_bin = mlb.fit_transform(y)
# direct transformation
[[1 1 0 0 0]
[1 0 1 0 0]
[0 1 0 1 0]
[1 0 1 1 0]
[0 0 1 0 1]]
# prediction of your classifier
y_pred = np.array([[1, 1, 0, 0, 0],
[1, 0, 1, 0, 0],
[0, 1, 0, 1, 0],
[1, 0, 1, 0, 0],
[1, 0, 1, 0, 0]])
# inverting the binarized values to the original classes
y_inv = mlb.inverse_transform(y_pred)
# output
[(0, 1), (0, 2), (1, 3), (0, 2), (0, 2)]

Using numba to randomly sample possible combinations of categories

I am trying to speed up a function that randomly samples a number of records with the possible combinations of a number of categories for a number of records and ensures they are unique (i.e. let's assume there's 3 records, any of them can be either 0 or 1 and I want 10 random samples of unique possible combinations of records).
If I did not use numba, I might would do something like this:
import numpy as np
def myfunc(categories, NumberOfRecords, maxsamples):
return np.unique( np.random.choice(np.arange(categories), size=(maxsamples*10, NumberOfRecords), replace=True), axis=0 )[0:maxsamples]
Annoyingly, numba does not support axis in np.unique, so I can do something like this, but some of the records may turn out to be non-unique.
from numba import njit, int64
import numpy as np
#njit(int64[:,:](int64, int64, int64), cache=True)
def myfunc(categories, NumberOfRecords, maxsamples):
return np.random.choice(np.arange(categories), size=(maxsamples, NumberOfRecords), replace=True)
myfunc(categories=2, NumberOfRecords=3, maxsamples=10)
E.g. in one call (obviously there's some randomness here), I got the below (for which the indices 1 and 6, and 3 and 4, and 7 and 9 are identical rows):
array([[0, 1, 1],
[1, 1, 0],
[0, 1, 0],
[1, 0, 1],
[1, 0, 1],
[1, 1, 1],
[1, 1, 0],
[1, 0, 0],
[0, 0, 0],
[1, 0, 0]])
My questions are:
Is this something where I would even expect a speed up from numba?
If so, how can I get a unique rows (this seems rather difficult with numba, but presumably there's a way)?
Perhaps there's a way to get at this more efficiently (perhaps without creating more random samples than I need in the end)?
In the following, I don't use numba, but all the operations use vectorized numpy functions.
Each row of the result that you generate can be interpreted as an integer expressed in base N, where N is the number of categories. With that interpretation, what you want is to sample without replacement from the integers [0, 1, ... N**R-1], where R is the number of "records". You can use the choice function for that, with the argument replace=False. Once you have that, you need to convert the chosen integers to base N. For that, I use the function int2base, which is a pared down version of a function that I wrote in a different answer.
Here's the code:
import numpy as np
def int2base(x, base, ndigits):
# x = np.asarray(x) # Uncomment this line for general purpose use.
powers = base ** np.arange(ndigits)
digits = (x.reshape(x.shape + (1,)) // powers) % base
return digits
def makesample(ncategories, nrecords, nsamples, rng=None):
if rng is None:
rng = np.random.default_rng()
n = ncategories ** nrecords
choices = rng.choice(n, replace=False, size=nsamples)
return int2base(choices, ncategories, nrecords)
In makesample, I included the optional argument rng. It allows you to specify the object that holds the choice function. If not provided, it uses np.random.default_rng().
Example:
In [118]: makesample(2, 3, 6)
Out[118]:
array([[0, 1, 1],
[0, 0, 1],
[1, 0, 1],
[0, 0, 0],
[1, 1, 0],
[1, 1, 1]])
In [119]: makesample(5, 4, 12)
Out[119]:
array([[3, 4, 0, 1],
[2, 0, 2, 0],
[4, 2, 4, 3],
[0, 1, 0, 4],
[0, 2, 0, 1],
[1, 2, 0, 1],
[0, 3, 0, 4],
[3, 3, 0, 3],
[3, 4, 1, 4],
[2, 4, 1, 1],
[3, 4, 1, 0],
[1, 1, 4, 4]])
makesample will raise an exception if you ask for too many samples:
In [120]: makesample(2, 3, 10)
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-120-80044e78a60a> in <module>
----> 1 makesample(2, 3, 10)
~/code_snippets/python/numpy/random_samples_for_so_question.py in makesample(ncategories, nrecords, nsamples, rng)
17 rng = np.random.default_rng()
18 n = ncategories ** nrecords
---> 19 choices = rng.choice(n, replace=False, size=nsamples)
20 return int2base(choices, ncategories, nrecords)
_generator.pyx in numpy.random._generator.Generator.choice()
ValueError: Cannot take a larger sample than population when 'replace=False'

Filled with numpy.zero but getting nan instead

I have a Class that contains various methods with include the following:
def _doc_mean(self, doc):
doc_vector_values = []
for w in doc:
#print(w)
if w.lower().strip() in self._E:
Q = np.zeros((1, 200), dtype=np.float64) #this is a zero array for when a word doesnt have a vector representation in our pretrained embeddings
doc_vector_values.append(self._E.get(w, Q))
with warnings.catch_warnings():
warnings.simplefilter("ignore", category=RuntimeWarning)
return np.mean(np.array(doc_vector_values, dtype=np.float64), axis=0)
def fit(self, X, y=None):
return self
def transform(self, X):
return np.array([self._doc_mean(doc) for doc in X])
def fit_transform(self, X, y=None):
return self.fit(X).transform(X)
in _doc_mean, i compare w with the keys in a dictionary E_, if there is a match, then load the value of the key-value pair which contains a 1*200 vector into a list, if there is no match, then load numpy.zeros((1,200)) into a list. This list is now converted to an array and the mean calculated.
When i instantiate the class and fit-transform my 'doc' data
mc = MeanClass()
X_ = mc.fit_transform(doc)
X_ is of dtype "object" and the places where there was a mismatch was replaced with nan instead of numpy.zero.
This leads to multiple other problems in my code that i cant fix. What am i doing wrong?
EDIT:
The E_ dictionary looks like this :
{'hello': array([ 5.84850e-02, 6.20640e-02, ..... -2.08990e-02])
'good': array([ -4.80050e-02, 2.80610e-02, ..... -5.04991e-02])
while doc looks like this :
['hello', 'bye', 'good']
['good', 'bye', 'night']
Since you haven't given a [mcve], I'll create something simple:
In [125]: E_ = {'foo':np.arange(5), 'bar':np.arange(1,6), 'baz':np.arange(5,10)}
In [126]: doc = ['foo','bar','sub','baz','foo']
Now do the dictionary lookup:
In [127]: alist = []
In [128]: for w in doc:
...: alist.append(E_.get(w,np.zeros((1,5),int)))
...:
In [129]: alist
Out[129]:
[array([0, 1, 2, 3, 4]),
array([1, 2, 3, 4, 5]),
array([[0, 0, 0, 0, 0]]),
array([5, 6, 7, 8, 9]),
array([0, 1, 2, 3, 4])]
In [130]: np.array(alist)
Out[130]:
array([array([0, 1, 2, 3, 4]), array([1, 2, 3, 4, 5]),
array([[0, 0, 0, 0, 0]]), array([5, 6, 7, 8, 9]),
array([0, 1, 2, 3, 4])], dtype=object)
The arrays in E_ are all shape (5,). The 'fill' array is (1,5). Due to the mismatch in shapes, the Out[130] array is 1d object.
I think you are trying to avoid the 'fill' case, but you test w.lower().strip() in self._E, and then use w in the get. So you might get the Q value sometimes. I got it with the 'sub' string.
If instead I make the 'fill' be (5,):
In [131]: alist = []
In [132]: for w in doc:
...: alist.append(E_.get(w,np.zeros((5,),int)))
...:
In [133]: alist
Out[133]:
[array([0, 1, 2, 3, 4]),
array([1, 2, 3, 4, 5]),
array([0, 0, 0, 0, 0]),
array([5, 6, 7, 8, 9]),
array([0, 1, 2, 3, 4])]
In [134]: np.array(alist)
Out[134]:
array([[0, 1, 2, 3, 4],
[1, 2, 3, 4, 5],
[0, 0, 0, 0, 0],
[5, 6, 7, 8, 9],
[0, 1, 2, 3, 4]])
The result is a (n,5) numeric array.
I can take two different means. One is the mean across all words, with a value for each 'attribute'. The other is the mean for each word, which I could just as well have gotten by taking the mean in E_.
In [135]: np.mean(_, axis=0)
Out[135]: array([1.2, 2. , 2.8, 3.6, 4.4])
In [137]: np.mean(__, axis=1)
Out[137]: array([2., 3., 0., 7., 2.]) # mean for each 'word'
mean of the object array in Out[130]:
In [138]: np.mean(_130, axis=0)
Out[138]: array([[1, 2, 2, 3, 4]])
The result is (1,5) and looks like Out[135] truncated, but I'd have to dig a bit further to be sure.
Hopefully this gives you an idea of what to watch out for. And an idea of the kind of 'minimal reproducable concrete example' that we find most useful.

How to use tf.gather in batch?

I have a A = 10x1000 tensor and a B = 10x1000 index tensor. The tensor B has values between 0-999 and it's used to gather values from A (B[0,:] gathers from A[0,:], B[1,:] from A[1,:], etc...).
However, if I use tf.gather(A, B) I get an array of shape (10, 1000, 1000) when I'm expecting a 10x1000 tensor back. Any ideas how I could fix this?
EDIT
Let's say A= [[1, 2, 3],[4,5,6]] and B = [[0, 1, 1],[2,1,0]] What I want is to be able to sample A using the corresponding B. This should result in C = [[1, 2, 2],[6,5,4]].
Dimensions of tensors are known in advance.
First we 'unstack' both the parameters and indices (A and B respectively) along the first dimension. Then we apply tf.gather() such that rows of A correspond to the rows of B. Finally, we stack together the result.
import tensorflow as tf
import numpy as np
def custom_gather(a, b):
unstacked_a = tf.unstack(a, axis=0)
unstacked_b = tf.unstack(b, axis=0)
gathered = [tf.gather(x, y) for x, y in zip(unstacked_a, unstacked_b)]
return tf.stack(gathered, axis=0)
a = tf.convert_to_tensor(np.array([[1, 2, 3], [4, 5, 6]]), tf.float32)
b = tf.convert_to_tensor(np.array([[0, 1, 1], [2, 1, 0]]), dtype=tf.int32)
gathered = custom_gather(a, b)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
print(sess.run(gathered))
# [[1. 2. 2.]
# [6. 5. 4.]]
For you initial case with shapes 1000x10 we get:
a = tf.convert_to_tensor(np.random.normal(size=(10, 1000)), tf.float32)
b = tf.convert_to_tensor(np.random.randint(low=0, high=999, size=(10, 1000)), dtype=tf.int32)
gathered = custom_gather(a, b)
print(gathered.get_shape().as_list()) # [10, 1000]
Update
The first dimension is unknown (i.e. None)
The previous solution works only if the first dimension is known in advance. If the dimension is unknown we solve it as follows:
We stack together two tensors such that the rows of both tensors are stacked together:
# A = [[1, 2, 3], [4, 5, 6]] [[[1 2 3]
# ---> [0 1 1]]
# [[4 5 6]
# B = [[0, 1, 1], [2, 1, 0]] [2 1 0]]]
We iterate over the elements of this stacked tensor (which consists of stacked together rows of A and B) and using tf.map_fn() function we apply tf.gather().
We stack back the elements we get with tf.stack()
import tensorflow as tf
import numpy as np
def custom_gather_v2(a, b):
def apply_gather(x):
return tf.gather(x[0], tf.cast(x[1], tf.int32))
a = tf.cast(a, dtype=tf.float32)
b = tf.cast(b, dtype=tf.float32)
stacked = tf.stack([a, b], axis=1)
gathered = tf.map_fn(apply_gather, stacked)
return tf.stack(gathered, axis=0)
a = np.array([[1, 2, 3], [4, 5, 6]], dtype=np.float32)
b = np.array([[0, 1, 1], [2, 1, 0]], dtype=np.int32)
x = tf.placeholder(tf.float32, shape=(None, 3))
y = tf.placeholder(tf.int32, shape=(None, 3))
gathered = custom_gather_v2(x, y)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
print(sess.run(gathered, feed_dict={x:a, y:b}))
# [[1. 2. 2.]
# [6. 5. 4.]]
Use tf.gather with batch_dims=-1:
import numpy as np
import tensorflow as tf
rois = np.array([[1, 2, 3],[3, 2, 1]])
ind = np.array([[0, 2, 1, 1, 2, 0, 0, 1, 1, 2],
[0, 1, 2, 0, 2, 0, 1, 2, 2, 2]])
tf.gather(rois, ind, batch_dims=-1)
# output:
# <tf.Tensor: shape=(2, 10), dtype=int64, numpy=
# array([[1, 3, 2, 2, 3, 1, 1, 2, 2, 3],
# [3, 2, 1, 3, 1, 3, 2, 1, 1, 1]])>

Elements in a list are overwritten

I tried to program a function which creates the linear span of a list of independent vectors, but it seems that the last calculated vector overwrites all other elements. I'd be nice if someone could help me fixing it.
def span_generator(liste,n):
"""function to generate the span of a list of linear independent
vectors(in liste) in the n-dimensional vectorspace of a finite
field with characteristic 2, returns a list of all elements which
lie inside the span"""
results=[]
blank=[]
for i in range(n):
blank.append(0)
a=blank
if len(liste)>1:
listenwert=liste[-1]
liste.pop(-1)
values=span_generator(liste,n)
for i in range(2):
for j in range(len(values)):
for k in range(n):
a[k]=(i*listenwert[k]+values[j][k])%2
results.append(a)
else:
for i in range(2):
for j in range(n):
a[j]=(i*liste[0][j])
results.append(a)
print(results)
return results
print(span_generator([[1,0],[0,1]],2)) gives following results
[[1, 0], [1, 0]]
[[1, 1], [1, 1], [1, 1], [1, 1]]
[[1, 1], [1, 1], [1, 1], [1, 1]]
instead of the expected: [[0,0],[1,0],[0,1],[1,1]]
Edit: I tried to simplify the program with itertools.product, but it didn't solve the problem.
def span_generator(liste):
n=len(liste[0])
results=[]
coeff=list(itertools.product(range(2), repeat=n))
blank=[]
for i in range(n):
blank.append(0)
for i in range(len(coeff)):
a=blank
for j in range(len(coeff[0])):
for k in range(n):
a[k]=(a[k]+coeff[i][j]*liste[j][k])%2
results.append(a)
return results
Output: span_generator([[0,1],[1,0]])
[[0, 0], [0, 0], [0, 0], [0, 0]]
But it should give [[0,0],[0,1],[1,0],[1,1]]
Another example: span_generator([[0,1,1],[1,1,0]]) should give [[0,0,0],[0,1,1],[1,1,0],[1,0,1]] (2=0 since i'm calculating modulo 2)
Coefficients
You can use itertools.product to generate the coefficients:
n = len(liste[0])
coefficients = itertools.product(range(2), repeat=len(liste))
yields an iterator with this content:
[(0, 0), (0, 1), (1, 0), (1, 1)]
Linear combinations
You can then selectively multiply the results with the transpose of your liste (list(zip(*liste)))
for coeff in coefficients:
yield [sum((a * c) for a, c in zip(transpose[i], coeff)) for i in range(n)]
which take for each dimensionality (for i in range(n)) the sum of the products
def span_generator3(liste):
n = len(liste[0])
transpose = list(zip(*liste))
coefficients = itertools.product(range(2), repeat=len(liste))
for coeff in coefficients:
yield [sum((a * c) for a, c in zip(transpose[i], coeff)) % 2 for i in range(n)]
this produces an iterator. If you want the result in a list-form, just can list() on the iterator
Result
list(span_generator3([[1,2],[4,8]]))
output:
[[0, 0], [4, 8], [1, 2], [5, 10]]
Higher dimensions
list(sorted(span_generator3([[1,2, 4],[8, 16, 32], [64, 128, 256]])))
output:
[[0, 0, 0],
[1, 2, 4],
[8, 16, 32],
[9, 18, 36],
[64, 128, 256],
[65, 130, 260],
[72, 144, 288],
[73, 146, 292]]
Modulo 2
If you want the result modulo 2, that's just adding 2 characters in the right place
def span_generator3_mod2(liste):
n = len(liste[0])
transpose = list(zip(*liste))
coefficients = itertools.product(range(2), repeat=len(liste))
# print(list(itertools.product(range(2), repeat=len(liste))))
for coeff in coefficients:
yield [sum((a * c) for a, c in zip(transpose[i], coeff)) % 2 for i in range(n)]
list(span_generator3_mod2([[0,1,1],[1,1,0]])) gives
[[0, 0, 0], [1, 1, 0], [0, 1, 1], [1, 0, 1]]

Resources