Use [`Tensor.tolist()`](https://pytorch.org/docs/stable/tensors.html#torch.Tensor.tolist) e.g:

&gt;     &gt;&gt;&gt; import torch
&gt;     &gt;&gt;&gt; a = torch.randn(2, 2)
&gt;     &gt;&gt;&gt; a.tolist()
&gt;     [[0.012766935862600803, 0.5415473580360413],
&gt;      [-0.08909505605697632, 0.7729271650314331]]
&gt;     &gt;&gt;&gt; a[0,0].tolist()
&gt;     0.012766935862600803

To remove all dimensions of size `1`, use `a.squeeze().tolist()`.

Alternatively, if all but one dimension are of size `1` (or you wish to get a list of every element of the tensor) you may use [`a.flatten().tolist()`](https://pytorch.org/docs/stable/torch.html#torch.flatten).


Tensor to list:

    a_list  = embeddings.tolist()

list to Tensor:

    a_tensor = torch.Tensor(a_list).cuda()


**#@pyorch** tensor to list or strings

    in[]
    #bbox predictions
    boxes = predictions[:, :4]
    print(boxes)
    
    out[]
    
    tensor(54.97658) tensor(393.99637) tensor(225.55316) tensor(879.53503)
    tensor(670.91669) tensor(400.35202) tensor(810.) tensor(878.34045)
    tensor(219.87546) tensor(408.02075) tensor(346.14133) tensor(860.66687)
    tensor(13.24882) tensor(217.30855) tensor(800.23413) tensor(737.75751)
    tensor(0.12453) tensor(552.29401) tensor(76.41209) tensor(885.35455)
    tensor(656.11823) tensor(625.61261) tensor(689.63586) tensor(713.37250)



after 

    in[]

    boxes = boxes.tolist()
    print(boxes)
        
    out []
        
    54.97657775878906 393.9963684082031 225.55316162109375 879.5350341796875
    670.9166870117188 400.3520202636719 810.0 878.3404541015625
    219.87545776367188 408.020751953125 346.1413269042969 860.6668701171875
    13.24881649017334 217.3085479736328 800.234130859375 737.7575073242188
    0.12453281879425049 552.2940063476562 76.4120864868164 885.3545532226562
    656.1182250976562 625.6126098632812 689.6358642578125 713.3724975585938




I have a few different interfaces and objects that each have a `type` property. Let&#39;s say these are objects stored in a NoSQL db. How can I create a generic `getItem` function with a deterministic return type based on its input parameter `type`? 

    interface Circle {
        type: &quot;circle&quot;;
        radius: number;
    }
    
    interface Square {
        type: &quot;square&quot;;
        length: number;
    }
    
    const shapes: (Circle | Square)[] = [
        { type: &quot;circle&quot;, radius: 1 },
        { type: &quot;circle&quot;, radius: 2 },
        { type: &quot;square&quot;, length: 10 }];
    
    function getItems(type: &quot;circle&quot; | &quot;square&quot;) {
        return shapes.filter(s =&gt; s.type == type);
        // Think of this as items coming from a database
        // I&#39;d like the return type of this function to be
        // deterministic based on the `type` value provided as a parameter. 
    }
    
    const circles = getItems(&quot;circle&quot;);
    for (const circle of circles) {
        console.log(circle.radius);
                           ^^^^^^
    }

&gt; Property &#39;radius&#39; does not exist on type &#39;Circle | Square&#39;.

TypeScript function return type based on input parameter

1. How can I perform aggregation with Pandas?
2. No DataFrame after aggregation! What happened?
3. How can I aggregate mainly strings columns (to `list`s, `tuple`s, `strings with separator`)?
4. How can I aggregate counts?
5. How can I create a new column filled by aggregated values?

I&#39;ve seen these recurring questions asking about various faces of the pandas aggregate functionality.
Most of the information regarding aggregation and its various use cases today is fragmented across dozens of badly worded, unsearchable posts.
The aim here is to collate some of the more important points for posterity.

This Q&amp;A is meant to be the next instalment in a series of helpful user-guides:

 - [How to pivot a dataframe][1],
 - [Pandas concat][2]
 - [How do I operate on a DataFrame with a Series for every column?][3]
 - [Pandas Merging 101][4]

Please note that this post is not meant to be a replacement for the [documentation about aggregation][5] and about [groupby][6], so please read that as well!

  [1]: https://stackoverflow.com/questions/47152691
  [2]: https://stackoverflow.com/questions/49620538
  [3]: https://stackoverflow.com/questions/53217607
  [4]: https://stackoverflow.com/questions/53645882
  [5]: http://pandas.pydata.org/pandas-docs/stable/groupby.html#aggregation
  [6]: http://pandas.pydata.org/pandas-docs/stable/groupby.html







Aggregation in Pandas

How do I convert a PyTorch `Tensor` into a python `list`?

I want to convert a tensor of size `[1, 2048, 1, 1]` into a list of 2048 elements. My tensor has floating point values. Is there a solution which also works with other data types such as int?

Convert PyTorch tensor to python list

<p>How do I convert a PyTorch <code>Tensor</code> into a python <code>list</code>?</p>
<p>I want to convert a tensor of size <code>[1, 2048, 1, 1]</code> into a list of 2048 elements. My tensor has floating point values. Is there a solution which also works with other data types such as int?</p>


**What are the most common pandas ways to select/filter rows of a [dataframe whose index is a MultiIndex](https://pandas.pydata.org/pandas-docs/stable/user_guide/advanced.html#)?**

 - Slicing based on a single value/label
 - Slicing based on multiple labels from one or more levels
 - Filtering on boolean conditions and expressions
 - Which methods are applicable in what circumstances

Assumptions for simplicity:

 1. input dataframe does not have duplicate index keys
 1. input dataframe below only has two levels. (Most solutions shown here generalize to N levels)

----


### Example input:
 
&gt;     mux = pd.MultiIndex.from_arrays([
&gt;         list(&#39;aaaabbbbbccddddd&#39;),
&gt;         list(&#39;tuvwtuvwtuvwtuvw&#39;)
&gt;     ], names=[&#39;one&#39;, &#39;two&#39;])
&gt;     
&gt;     df = pd.DataFrame({&#39;col&#39;: np.arange(len(mux))}, mux)
&gt;     
&gt;              col
&gt;     one two     
&gt;     a   t      0
&gt;         u      1
&gt;         v      2
&gt;         w      3
&gt;     b   t      4
&gt;         u      5
&gt;         v      6
&gt;         w      7
&gt;         t      8
&gt;     c   u      9
&gt;         v     10
&gt;     d   w     11
&gt;         t     12
&gt;         u     13
&gt;         v     14
&gt;         w     15

## Question 1: Selecting a Single Item       
How do I select rows having &quot;a&quot; in level &quot;one&quot;? 

             col
    one two     
    a   t      0
        u      1
        v      2
        w      3

Additionally, how would I be able to drop level &quot;one&quot; in the output?

         col
    two     
    t      0
    u      1
    v      2
    w      3


**Question 1b**    
How do I slice all rows with value &quot;t&quot; on level &quot;two&quot;?
 
             col
    one two     
    a   t      0
    b   t      4
        t      8
    d   t     12

## Question 2: Selecting Multiple Values in a Level        
How can I select rows corresponding to items &quot;b&quot; and &quot;d&quot; in level &quot;one&quot;?


             col
    one two     
    b   t      4
        u      5
        v      6
        w      7
        t      8
    d   w     11
        t     12
        u     13
        v     14
        w     15

**Question 2b**    
How would I get all values corresponding to &quot;t&quot; and &quot;w&quot; in level &quot;two&quot;?

             col
    one two     
    a   t      0
        w      3
    b   t      4
        w      7
        t      8
    d   w     11
        t     12
        w     15


## Question 3: Slicing a Single Cross Section `(x, y)`      
How do I retrieve a cross section, i.e., a single row having a specific values for the index from `df`? Specifically, how do I retrieve the cross section of `(&#39;c&#39;, &#39;u&#39;)`, given by

             col
    one two     
    c   u      9

## Question 4: Slicing Multiple Cross Sections `[(a, b), (c, d), ...]`       
How do I select the two rows corresponding to `(&#39;c&#39;, &#39;u&#39;)`, and `(&#39;a&#39;, &#39;w&#39;)`?

             col
    one two     
    c   u      9
    a   w      3

## Question 5: One Item Sliced per Level       
How can I retrieve all rows corresponding to &quot;a&quot; in level &quot;one&quot; or &quot;t&quot; in level &quot;two&quot;?

             col
    one two     
    a   t      0
        u      1
        v      2
        w      3
    b   t      4
        t      8
    d   t     12

## Question 6: Arbitrary Slicing    
How can I slice specific cross sections? For &quot;a&quot; and &quot;b&quot;, I would like to select all rows with sub-levels &quot;u&quot; and &quot;v&quot;, and for &quot;d&quot;, I would like to select rows with sub-level &quot;w&quot;.

             col
    one two     
    a   u      1
        v      2
    b   u      5
        v      6
    d   w     11
        w     15


&gt; Question 7 will use a unique setup consisting of a numeric level:
&gt; 
&gt;     np.random.seed(0)
&gt;     mux2 = pd.MultiIndex.from_arrays([
&gt;         list(&#39;aaaabbbbbccddddd&#39;),
&gt;         np.random.choice(10, size=16)
&gt;     ], names=[&#39;one&#39;, &#39;two&#39;])
&gt;     
&gt;     df2 = pd.DataFrame({&#39;col&#39;: np.arange(len(mux2))}, mux2)
&gt;     
&gt;              col
&gt;     one two     
&gt;     a   5      0
&gt;         0      1
&gt;         3      2
&gt;         3      3
&gt;     b   7      4
&gt;         9      5
&gt;         3      6
&gt;         5      7
&gt;         2      8
&gt;     c   4      9
&gt;         7     10
&gt;     d   6     11
&gt;         8     12
&gt;         8     13
&gt;         1     14
&gt;         6     15


## Question 7: Filtering by numeric inequality on individual levels of the multiindex      
How do I get all rows where values in level &quot;two&quot; are greater than 5?

             col
    one two     
    b   7      4
        9      5
    c   7     10
    d   6     11
        8     12
        8     13
        6     15

-----
Note: This post will *not* go through how to create MultiIndexes, how to perform assignment operations on them, or any performance related discussions (these are separate topics for another time). 

Select rows in pandas MultiIndex DataFrame

I am using the following setup

- [macOS v10.14][1] (Mojave)
- Python 3.7.1
- Visual Studio Code 1.30
- Pylint 2.2.2
- Django 2.1.4

I want to use linting to make my life a bit easier in Visual Studio Code. However, for every import I have states &quot;unresolved import&quot;. Even on default Django imports (i.e. from django.db import models).

I presume it is because it is not seeing the virtual environment Python files.

Everything works just fine, but it&#39;s starting to get annoying.

The interpreter choices I have are all system versions of Python. It does not seem to see my virtual environment Python at all (it is not in the same directory as my workspace, so that part makes sense).

If I set up the *python.PythonPath* in the *settings.json* file, it just ignores it and does not list my virtual environment path as an option. I also tried setting it up in my global Python settings, but it also does not show up.

Is there a quick fix to get it working?

  [1]: https://en.wikipedia.org/wiki/MacOS_Mojave






Pylint &quot;unresolved import&quot; error in Visual Studio Code

In Python 3.3 &quot;abstract base classes&quot; in `collections` (like `MutableMapping` or `MutableSequence`) were moved to second-level module `collections.abc`. So in Python 3.3+ the real type is `collections.abc.MutableMapping` and so on. [Documentation][1] states that the old alias names (e.g. `collections.MutableMapping`) will be available up to Python 3.7 (currently the latest version), however in 3.8 these aliases will be removed.

Current version of Python 3.7 even produces a warning when you use the alias names:

    ./scripts/generateBoard.py:145: DeprecationWarning: Using or importing the ABCs from &#39;collections&#39; instead of from &#39;collections.abc&#39; is deprecated, and in 3.8 it will stop working
      elif isinstance(value, (collections.MutableMapping, collections.MutableSequence)) == True:

In python 2.7 there is no `collections.abc`.

How can Python script handle this difference in the most convenient way, when it is meant to be used with (almost) any Python version? I&#39;m looking for a solution which would ideally solve this mess in one central place, without having to use `try: ... except: ...` all over the script everywhere I need this type?


  [1]: https://docs.python.org/3/library/collections.html

How to use collections.abc from both Python 3.8+ and Python 2.7

If you execute the following statement in Python 3.7, it will (from my testing) print `b`:

    if None.__eq__(&quot;a&quot;):
        print(&quot;b&quot;)

However, `None.__eq__(&quot;a&quot;)` evaluates to `NotImplemented`.

Naturally, `&quot;a&quot;.__eq__(&quot;a&quot;)` evaluates to `True`, and `&quot;b&quot;.__eq__(&quot;a&quot;)` evaluates to `False`.

I initially discovered this when testing the return value of a function, but didn&#39;t return anything in the second case -- so, the function returned `None`.

What&#39;s going on here?

Why does `if None.__eq__(&quot;a&quot;)` seem to evaluate to True (but not quite)?

1. If `num_workers` is 2, Does that mean that it will put 2 batches in the RAM and send 1 of them to the GPU or Does it put 3 batches in the RAM then sends 1 of them to the GPU?
1. What does actually happen when the number of workers is higher than the number of CPU cores? I tried it and it worked fine but How does it work? (I thought that the maximum number of workers I can choose is the number of cores).
1. If I set `num_workers` to 3 and during the training there were no batches in the memory for the GPU, Does the main process waits for its workers to read the batches or Does it read a single batch (without waiting for the workers)?

How does the &quot;number of workers&quot; parameter in PyTorch dataloader actually work?

Where is an explicit connection between the `optimizer` and the `loss`? 

How does the optimizer know where to get the gradients of the loss without a call liks this `optimizer.step(loss)`?

-More context-

When I minimize the loss, I didn&#39;t have to pass the gradients to the optimizer.

    loss.backward() # Back Propagation
    optimizer.step() # Gardient Descent



pytorch - connection between loss.backward() and optimizer.step()

What&#39;s the difference between [`torch.cat`][3] and [`torch.stack`][4]?

---

OpenAI&#39;s [REINFORCE][1] and [actor-critic][2] examples for reinforcement learning have the following:

    # REINFORCE:
    policy_loss = torch.cat(policy_loss).sum()

    # actor-critic:
    loss = torch.stack(policy_losses).sum() + torch.stack(value_losses).sum()



  [1]: https://github.com/pytorch/examples/blob/master/reinforcement_learning/reinforce.py
  [2]: https://github.com/pytorch/examples/blob/master/reinforcement_learning/actor_critic.py
  [3]: https://pytorch.org/docs/stable/generated/torch.cat.html
  [4]: https://pytorch.org/docs/stable/generated/torch.stack.html

What&#39;s the difference between torch.stack() and torch.cat()?

I successfully trained the network but got this error during validation:

&gt; RuntimeError: CUDA error: out of memory

How to fix this strange error: &quot;RuntimeError: CUDA error: out of memory&quot;

What is the correct way to perform gradient clipping in pytorch?

I have an exploding gradients problem.

Content Type	Original Author	Original Content on Stackoverflow
Question	Tom Hale	View Question on Stackoverflow
Solution 1 - Python	Tom Hale	View Answer on Stackoverflow
Solution 2 - Python	Wesam Na	View Answer on Stackoverflow
Solution 3 - Python	Bharath Kumar	View Answer on Stackoverflow

Convert PyTorch tensor to python list

Python Problem Overview

Python Solutions

Solution 1 - Python

Solution 2 - Python

Solution 3 - Python

Aggregation in Pandas

TypeScript function return type based on input parameter

Attributions