Error when broadcasting to numpy matrix - python

I know this is a relatively common topic on stackoverflow but I couldn't find the answer I was looking for. Basically, I am trying to make very efficient code (I have rather large data sets) to get certain columns of data from a matrix. Below is what I have so far. It gives me this error: could not broadcast input array from shape (2947,1) into shape (2947)
def get_data(self, colHeaders):
temp = np.zeros((self.matrix_data.shape[0],len(colHeaders)))
for col in colHeaders:
index = self.header2matrix[col]
temp[:,index:] = self.matrix_data[:,index]
data = np.matrix(temp)
return temp

Maybe this simple example will help:
In [70]: data=np.arange(12).reshape(3,4)
In [71]: header={'a':0,'b':1,'c':2}
In [72]: col=['c','a']
In [73]: index=[header[i] for i in col]
In [74]: index
Out[74]: [2, 0]
In [75]: data[:,index]
Out[75]:
array([[ 2, 0],
[ 6, 4],
[10, 8]])
data is some sort of 2D array, header is a dictionary mapping names to column numbers. Using the input col, I construct a column index list. You can select all columns at once, rather than one by one.

Related

Iterating over a pandas sparse series, without the missing values

I have a pandas DataFrame with very sparse columns. I would like to iterate over the DataFrame's values but without the missing ones, to save time.
I can't find how to access the indexes of the non-empty cells.
For example:
a = pd.Series([2, 3, 0, 0, 4], dtype='Sparse[int]')
print(a.sparse.sp_values) # --> [2,3,4]
print(a.sparse.sp_index) # --> AttributeError
print(a.sparse.to_coo()) # --> ValueError
I got the non-empty values, but where is the index? In the above example I am looking for [0,1,4].
I looked at the documentation which doesn't seem to mention it. I found information only for SparseArray but not for a Series/DataFrame of sparse type.
Printing dir(a.sparse) (without those starting with '_'):
['density', 'fill_value', 'from_coo', 'npoints', 'sp_values', 'to_coo', 'to_dense']
IIUC, use flatnonzero from numpy :
idx = np.flatnonzero(a).tolist()
print(idx)
​#[0, 1, 4]
Or loc from pandas's with boolean indexing :
idx = a.ne(0).loc[lambda s: s].index.tolist() # or list(a[a.ne(0)].index)
print(idx)
#[0, 1, 4]

Numpy: proper way to arrange input and output vectors

I am trying to create a numpy array with 2 columns and multiple rows. The first column is meant to represent input vector of size 3. The 2nd column is meant to represent output vector of size 2.
arr = np.array([
[np.array([1,2,3]), np.array([1,0])]
[np.array([4,5,6]), np.array([0,1])]
])
I was expecting: arr[:, 0].shape
to return (2, 3), but it returns (2, )
What is the proper way to arrange input and output vectors into a matrix using numpy?
If you are sure the elements in each column have the same size/length, you can select and then stack the result using numpy.row_stack:
np.row_stack(arr[:,0]).shape
# (2, 3)
np.row_stack(arr[:,1]).shape
# (2, 2)
So, the code
arr = np.array([
[np.array([1,2,3]), np.array([1,0])],
[np.array([4,5,6]), np.array([0,1])]
])
Creates an object array, indexing the first column gives you back two rows with one object in each, which accounts for the size. To get what you want you'd need to wrap it in something like
np.vstack(arr[:, 0])
Which creates an array out of the objects in the first column. This isn't very convenient, it would make more sense to me to store these in a dictionary, something like
io = {'in': np.array([[1,2,3],[4,5,6]]),
'out':np.array([[1,0], [0,1]])
}
A structured array gives you a bit of both. Creation is a bit tricky, for the example given,
arr = np.array([
(1,2,3), (1,0)),
((4,5,6), (0,1)) ],
dtype=[('in', '3int64'), ('out', '2float64')])
Creates a structured array with fields in and out, consisting of 3 integers and 2 floats respectively. Rows can be accessed as usual,
In[73]: arr[0]
Out[74]: ([1, 2, 3], [ 1., 0.])
Or by the field name
In [73]: arr['in']
Out[73]:
array([[1, 2, 3],
[4, 5, 6]])
The numpy manual has many more details (https://docs.scipy.org/doc/numpy-1.13.0/user/basics.rec.html). I can't add any details as I've been intending to use them in a project for some time, but haven't.

Reduce size of array based on multiple column criteria in python

I need to reduce the size of an array, based on criteria found on another array; I need to look into the relationships and change the value based on the new information. Here is a simplified version of my problem.
I have an array (or dataframe) with my data:
data = np.array([[[[1, 2, 3, 4], [5, 6, 7, 8]]]]).reshape((4,2))
I have another file, of different size, that holds information about the values in the data array:
a = np.array([[1, 1, 2],[2, 3, 4],[3, 5, 6], [4, 7, 8] ]).reshape((4,3))
The information I have in a tells me how I can reduce the size of data, for example a[0] tells me that data[0][0:2] == a[0][1:].
so I can replace the unique value a[0][0:1] with data[0][0:2] (effectively reducing the size of array data
To clarify, array a holds three pieces of information per position, a[0] has the information 1, 1, 2 - now I want to scan through the data array, and when the a[i][1:] is equal to any of the data[i][0:2] or data[i][2:] then I want to replace the value with the a[i][0:1] - is that any clearer?
my final array should be like this:
new_format = np.array([[[[1, 2], [3,4]]]]).reshape((2,2))
There are questions like the following: Filtering a DataFrame based on multiple column criteria
but are only based on filtering based on certain numerical criteria.
I figured out a way to do it, using the pandas library. Probably not the best solution, but worked from me.
In my case I read the data in the pandas library, but for the posted example I can convert the arrays to dataframes
datas = pd.DataFrame(data) ##convert to dataframe
az = pd.DataFrame(a)
datas= datas.rename(columns={'0': '1', '1': '2'}) ## rename columns for comparison with a array
new_format= pd.merge(datas, az, how='right') #do the comparison
new_format = new_format.drop(['1','2'],1) #drop the old columns, keeping only the new format

Access Columns of Numpy Array? Errors Trying to Do by Transpose or by Column Access

I've a numpy.ndarray the columns of which I'd like to access. I will be taking all columns after 8 and testing them for variance, removing the column if the variance/average is low. In order to do this, I need access to the columns, preferably with Numpy. By my current methods, I encounter errors or failure to transpose.
To mine these arrays, I am using the IOPro adapter, which gives a regular numpy.ndarray.
import iopro
import sys
adapter = iopro.text_adapter(sys.argv[1], parser='csv')
all_data = adapter[:]
z_matrix = adapter[range(8,len(all_data[0]))][1:3]
print type(z_matrix) #check type
print z_matrix # print array
print z_matrix.transpose() # attempt transpose (fails)
print z_matrix[:,0] # attempt access by column (fails)
Can someone explain what is happening?
The output is this:
<type 'numpy.ndarray'>
[ (18.712, 64.903, -10.205, -1.346, 0.319, -0.654, 1.52398, 114.495, -75.2488, 1.52184, 111.31, 175.
408, 1.52256, 111.699, -128.141, 1.49227, 111.985, -138.173)
(17.679, 48.015, -3.152, 0.848, 1.239, -0.3, 1.52975, 113.963, -50.0622, 1.52708, 112.335, -57.4621
, 1.52603, 111.685, -161.098, 1.49204, 113.406, -66.5854)]
[ (18.712, 64.903, -10.205, -1.346, 0.319, -0.654, 1.52398, 114.495, -75.2488, 1.52184, 111.31, 175.
408, 1.52256, 111.699, -128.141, 1.49227, 111.985, -138.173)
(17.679, 48.015, -3.152, 0.848, 1.239, -0.3, 1.52975, 113.963, -50.0622, 1.52708, 112.335, -57.4621
, 1.52603, 111.685, -161.098, 1.49204, 113.406, -66.5854)]
Traceback (most recent call last):
File "z-matrix-filtering.py", line 11, in <module>
print z_matrix[:,0]
IndexError: too many indices
What is going wrong? Is there a better way to access the columns? I will be reading all lines of a file, testing all columns from the 8th for significant variance, removing any columns that don't vary significantly, and then reprinting the result as a new CSV.
EDIT:
Based on responses, I have created the following very ugly and I think inane approach.
all_data = adapter[:]
z_matrix = []
for line in all_data:
to_append = []
for column in range(8,len(all_data.dtype)):
to_append.append(line[column].astype(np.float16))
z_matrix.append(to_append)
z_matrix = np.array(z_matrix)
The reason that the columns must be directly accessed is that there is a String inside the data. If this string is not circumvented in some way, an error will be thrown about a void-array with object members using buffer error.
Is there a better solution? This seems terrible, and it seems it will be inefficient for several gigabytes of data.
Notice that the output of print z_matrix has the form
[ (18.712, 64.903, ..., -138.173)
(17.679, 48.015, ..., -66.5854)]
That is, it is printed as a list of tuples. That is the output you get when the array is a "structured array". It is a one-dimensional array of structures. Each "element" in the array has 18 fields. The error occurs because you are trying to index a 1-D array as if it were 2-D; z_matrix[:,0] won't work.
Print the data type of the array to see the details. E.g.
print z_matrix.dtype
That should show the names of the fields and their individual data types.
You can get one of the elements as, for example, z_matrix[k] (where k is an integer), or you can access a "column" (really a field of the structured array) as z_matrix['name'] (change 'name' to one of the fields in the dtype).
If the fields all have the same data type (which looks like the case here--each field has type np.float64), you can create a 2-D view of the data by reshaping the result of the view method. For example:
z_2d = z_matrix.view(np.float64).reshape(-1, len(z_matrix.dtype.names))
Another way to get the data by column number rather than name is:
col = 8 # The column number (zero-based).
col_data = z_matrix[z_matrix.dtype.names[col]]
For more about structured arrays, see http://docs.scipy.org/doc/numpy/user/basics.rec.html.
The display of z_matrix is consistent with it being shape (2,), a 1d array of tuples.
np.array([np.array(a) for a in z_matrix])
produces a (2,18) 2d array. You should be able to do your column tests on that.
It is very easy to access numpy array. Here's a simple example which can be helpful
import numpy as n
A = n.array([[1, 2, 3], [4, 5, 6]])
print A
>>> array([[1, 2, 3],
[5, 6, 7]])
A.T // To obtain the transpose
>>> array([[1, 5],
[2, 6],
[3, 7]])
n.mean(A.T, axis = 1) // To obtain column wise mean of array A
>>> array([ 3., 4., 5.])
I hope this will help you perform your transpose and column-wise operations

Slicing of a NumPy 2d array, or how do I extract an mxm submatrix from an nxn array (n>m)?

I want to slice a NumPy nxn array. I want to extract an arbitrary selection of m rows and columns of that array (i.e. without any pattern in the numbers of rows/columns), making it a new, mxm array. For this example let us say the array is 4x4 and I want to extract a 2x2 array from it.
Here is our array:
from numpy import *
x = range(16)
x = reshape(x,(4,4))
print x
[[ 0 1 2 3]
[ 4 5 6 7]
[ 8 9 10 11]
[12 13 14 15]]
The line and columns to remove are the same. The easiest case is when I want to extract a 2x2 submatrix that is at the beginning or at the end, i.e. :
In [33]: x[0:2,0:2]
Out[33]:
array([[0, 1],
[4, 5]])
In [34]: x[2:,2:]
Out[34]:
array([[10, 11],
[14, 15]])
But what if I need to remove another mixture of rows/columns? What if I need to remove the first and third lines/rows, thus extracting the submatrix [[5,7],[13,15]]? There can be any composition of rows/lines. I read somewhere that I just need to index my array using arrays/lists of indices for both rows and columns, but that doesn't seem to work:
In [35]: x[[1,3],[1,3]]
Out[35]: array([ 5, 15])
I found one way, which is:
In [61]: x[[1,3]][:,[1,3]]
Out[61]:
array([[ 5, 7],
[13, 15]])
First issue with this is that it is hardly readable, although I can live with that. If someone has a better solution, I'd certainly like to hear it.
Other thing is I read on a forum that indexing arrays with arrays forces NumPy to make a copy of the desired array, thus when treating with large arrays this could become a problem. Why is that so / how does this mechanism work?
To answer this question, we have to look at how indexing a multidimensional array works in Numpy. Let's first say you have the array x from your question. The buffer assigned to x will contain 16 ascending integers from 0 to 15. If you access one element, say x[i,j], NumPy has to figure out the memory location of this element relative to the beginning of the buffer. This is done by calculating in effect i*x.shape[1]+j (and multiplying with the size of an int to get an actual memory offset).
If you extract a subarray by basic slicing like y = x[0:2,0:2], the resulting object will share the underlying buffer with x. But what happens if you acces y[i,j]? NumPy can't use i*y.shape[1]+j to calculate the offset into the array, because the data belonging to y is not consecutive in memory.
NumPy solves this problem by introducing strides. When calculating the memory offset for accessing x[i,j], what is actually calculated is i*x.strides[0]+j*x.strides[1] (and this already includes the factor for the size of an int):
x.strides
(16, 4)
When y is extracted like above, NumPy does not create a new buffer, but it does create a new array object referencing the same buffer (otherwise y would just be equal to x.) The new array object will have a different shape then x and maybe a different starting offset into the buffer, but will share the strides with x (in this case at least):
y.shape
(2,2)
y.strides
(16, 4)
This way, computing the memory offset for y[i,j] will yield the correct result.
But what should NumPy do for something like z=x[[1,3]]? The strides mechanism won't allow correct indexing if the original buffer is used for z. NumPy theoretically could add some more sophisticated mechanism than the strides, but this would make element access relatively expensive, somehow defying the whole idea of an array. In addition, a view wouldn't be a really lightweight object anymore.
This is covered in depth in the NumPy documentation on indexing.
Oh, and nearly forgot about your actual question: Here is how to make the indexing with multiple lists work as expected:
x[[[1],[3]],[1,3]]
This is because the index arrays are broadcasted to a common shape.
Of course, for this particular example, you can also make do with basic slicing:
x[1::2, 1::2]
As Sven mentioned, x[[[0],[2]],[1,3]] will give back the 0 and 2 rows that match with the 1 and 3 columns while x[[0,2],[1,3]] will return the values x[0,1] and x[2,3] in an array.
There is a helpful function for doing the first example I gave, numpy.ix_. You can do the same thing as my first example with x[numpy.ix_([0,2],[1,3])]. This can save you from having to enter in all of those extra brackets.
I don't think that x[[1,3]][:,[1,3]] is hardly readable. If you want to be more clear on your intent, you can do:
a[[1,3],:][:,[1,3]]
I am not an expert in slicing but typically, if you try to slice into an array and the values are continuous, you get back a view where the stride value is changed.
e.g. In your inputs 33 and 34, although you get a 2x2 array, the stride is 4. Thus, when you index the next row, the pointer moves to the correct position in memory.
Clearly, this mechanism doesn't carry well into the case of an array of indices. Hence, numpy will have to make the copy. After all, many other matrix math function relies on size, stride and continuous memory allocation.
If you want to skip every other row and every other column, then you can do it with basic slicing:
In [49]: x=np.arange(16).reshape((4,4))
In [50]: x[1:4:2,1:4:2]
Out[50]:
array([[ 5, 7],
[13, 15]])
This returns a view, not a copy of your array.
In [51]: y=x[1:4:2,1:4:2]
In [52]: y[0,0]=100
In [53]: x # <---- Notice x[1,1] has changed
Out[53]:
array([[ 0, 1, 2, 3],
[ 4, 100, 6, 7],
[ 8, 9, 10, 11],
[ 12, 13, 14, 15]])
while z=x[(1,3),:][:,(1,3)] uses advanced indexing and thus returns a copy:
In [58]: x=np.arange(16).reshape((4,4))
In [59]: z=x[(1,3),:][:,(1,3)]
In [60]: z
Out[60]:
array([[ 5, 7],
[13, 15]])
In [61]: z[0,0]=0
Note that x is unchanged:
In [62]: x
Out[62]:
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11],
[12, 13, 14, 15]])
If you wish to select arbitrary rows and columns, then you can't use basic slicing. You'll have to use advanced indexing, using something like x[rows,:][:,columns], where rows and columns are sequences. This of course is going to give you a copy, not a view, of your original array. This is as one should expect, since a numpy array uses contiguous memory (with constant strides), and there would be no way to generate a view with arbitrary rows and columns (since that would require non-constant strides).
With numpy, you can pass a slice for each component of the index - so, your x[0:2,0:2] example above works.
If you just want to evenly skip columns or rows, you can pass slices with three components
(i.e. start, stop, step).
Again, for your example above:
>>> x[1:4:2, 1:4:2]
array([[ 5, 7],
[13, 15]])
Which is basically: slice in the first dimension, with start at index 1, stop when index is equal or greater than 4, and add 2 to the index in each pass. The same for the second dimension. Again: this only works for constant steps.
The syntax you got to do something quite different internally - what x[[1,3]][:,[1,3]] actually does is create a new array including only rows 1 and 3 from the original array (done with the x[[1,3]] part), and then re-slice that - creating a third array - including only columns 1 and 3 of the previous array.
I have a similar question here: Writting in sub-ndarray of a ndarray in the most pythonian way. Python 2
.
Following the solution of previous post for your case the solution looks like:
columns_to_keep = [1,3]
rows_to_keep = [1,3]
An using ix_:
x[np.ix_(rows_to_keep, columns_to_keep)]
Which is:
array([[ 5, 7],
[13, 15]])
I'm not sure how efficient this is but you can use range() to slice in both axis
x=np.arange(16).reshape((4,4))
x[range(1,3), :][:,range(1,3)]

Categories