# Reading NDTensor.BlockSparse

**URL:** <https://itensor.discourse.group/t/reading-ndtensor-blocksparse/1516>\
**Category:** ITensor Julia Questions\
**Created:** [March 1, 2024, 8:50am UTC](https://itensor.discourse.group/t/reading-ndtensor-blocksparse/1516 "2024-03-01T08:50:05Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![FTarocco](https://avatars.discourse-cdn.com/v4/letter/f/85f322/32.png) [@FTarocco](https://itensor.discourse.group/u/FTarocco)\
**Post date:** [March 1, 2024, 8:50am UTC](https://itensor.discourse.group/t/reading-ndtensor-blocksparse/1516/1 "2024-03-01T08:50:05Z")

</div>

Hi,  
I’m trying to import an MPS in HDF5 format in python.  
Of all the data contained in the file, I’ve extracted only the components that I needed to read the values inside the MPS.  
Each mps tensor, contains a vector of float which are the value different from 0, and an offsets object that give information about in which block the data is not 0.

```python
{'data': array([ 9.99233978e-01, -3.91338417e-02, -6.12386558e-10, -8.88178420e-15,
         3.22253884e-06, 3.89747029e-09, -2.49062916e-01, 2.16631132e-06,
         6.16173779e-15, 8.57122151e-11, -5.47718360e-03, -1.16275436e-01,
        -3.23899185e-10, -2.51718972e-01, -3.93968386e-09, 1.58006064e-09,
        -5.67219778e-01, 9.68487168e-01, 4.50166647e-04, -2.57321454e-04,
         2.94209102e-15, 1.79324504e-05, 1.76508249e-15, -1.01415632e-03,
        -9.66351958e-01, -1.99233940e-01, 8.31695823e-14, 1.13660573e-01,
        -8.82614208e-16, -1.64590563e-13, -5.45195779e-13, -3.21726803e-13,
        -7.84155140e-01, -4.65105732e-12, -4.49876820e-11, 3.89411336e-01,
        -8.78305991e-01, 2.16576571e-01, -2.67414464e-14, -6.32869735e-02,
        -1.44138174e-14, 3.74064820e-01, 3.07336758e-01, 8.17120438e-01,
        -2.44312687e-13, 1.12015076e-01, 1.64625258e-14, -1.65344821e-14,
        -1.47631126e-14, 3.58094190e-15, -2.77674267e-11, -5.57448322e-15,
         4.84002892e-01, 3.76647383e-10, -1.60041481e-01, -2.03339403e-02,
        -3.64485763e-10, -6.81958705e-10, 2.89771192e-01, -3.76709004e-02,
        -6.75240161e-10, 8.75065464e-01, 2.05941465e-09, 2.29161135e-11,
        -1.27834978e-03, 2.10775034e-02, 4.96030312e-11, -4.27827354e-09,
         2.38678051e-01, 9.70869984e-01]),
 'offset': [([2, 1, 1], 0),
  ([1, 2, 1], 1),
  ([2, 2, 2], 5),
  ([1, 3, 2], 17),
  ([2, 3, 3], 35),
  ([1, 4, 3], 53),
  ([2, 4, 4], 65),
  ([1, 5, 4], 69)],
 'n_site': 4,
 'tag': ['n=4', 'l=4', 'l=3'],
 'dims': [2, 16, 8]}

```

If i want to reconstruct whole tensor in python, starting from a np.zeros(dims) , where I can find the information about which index or indices I have to iterate over to insert the values?

Otherwise, I can directly convert each mps tensor into array(ITensor(mps[i])), save it into a JSON file, and read it from python changing the construction order of a np.array from row major to column major.

---

<div class="post-metadata">

**Author:** ![mtfishman](https://yyz2.discourse-cdn.com/free1/user_avatar/itensor.discourse.group/mtfishman/32/11_2.png) [@mtfishman](https://itensor.discourse.group/u/mtfishman)\
**Post date:** [March 1, 2024, 1:43pm UTC](https://itensor.discourse.group/t/reading-ndtensor-blocksparse/1516/2 "2024-03-01T13:43:19Z")

</div>

You will also need the block structure of the indices to know how big the blocks are.

---

<div class="post-metadata">

**Author:** ![FTarocco](https://avatars.discourse-cdn.com/v4/letter/f/85f322/32.png) [@FTarocco](https://itensor.discourse.group/u/FTarocco)\
**Post date:** [March 1, 2024, 2:12pm UTC](https://itensor.discourse.group/t/reading-ndtensor-blocksparse/1516/3 "2024-03-01T14:12:32Z")

</div>

For block structure, you refer to the second row, right?  
 ![image](https://global.discourse-cdn.com/free1/uploads/itensor/original/1X/a68421ee43a6e8e82654d0fb9fdd3db22d26a7c5.png)

In which subsection I can find the block structure?

 ![image](https://global.discourse-cdn.com/free1/uploads/itensor/original/1X/8577ac83bdf5b6e163375ff3e13a968143313c33.png)

---

<div class="post-metadata">

**Author:** ![mtfishman](https://yyz2.discourse-cdn.com/free1/user_avatar/itensor.discourse.group/mtfishman/32/11_2.png) [@mtfishman](https://itensor.discourse.group/u/mtfishman)\
**Post date:** [March 1, 2024, 3:04pm UTC](https://itensor.discourse.group/t/reading-ndtensor-blocksparse/1516/4 "2024-03-01T15:04:42Z")

</div>

The `space` directory should have the block structure of each index/dimension of the tensor.