`JPG` does not support *transparency* - `RGBA` means `Red`, `Green`, `Blue`, `Alpha` - `Alpha` is *transparency*. 

You need to discard the `Alpha` Channel or save as something that supports transparency - like `PNG`.

The `Image` class has a method `convert` which can be used to convert `RGBA` to `RGB` - after that you will be able to save as `JPG`.

Have a look here: [the image class doku](http://pillow.readthedocs.io/en/3.1.x/reference/Image.html#the-image-class)


    im = Image.open(&quot;audacious.png&quot;)
    rgb_im = im.convert(&#39;RGB&#39;)
    rgb_im.save(&#39;audacious.jpg&#39;)

Adapted from dm2013&#39;s answer to [Convert png to jpeg using Pillow](https://stackoverflow.com/a/43258974/7505395)

There was an error when I tried to remove a package, so it was left in a broken state. I tried the following:

    sudo dpkg --remove --force-remove-reinstreq rvm

Output:

    (Reading database ... 41743 files and directories currently installed.)
    Removing rvm (1.29.3-1) ...
    bash: rvm: command not found
    dpkg: error processing package rvm (--purge):
     subprocess installed pre-removal script returned error exit status 127
    Errors were encountered while processing:
     rvm
.

    sudo dpkg --purge rvm

Output: same as above

    sudo dpkg -i --force-overwrite package

Output:

    dpkg: error processing archive rvm (--install):
     cannot access archive: No such file or directory
    Errors were encountered while processing:
     rvm

Is there a way to force remove it, ignoring all errors? I&#39;m using Ubuntu 16.04 on WSL.

Removing broken packages in Ubuntu

I&#39;m automating my Docker installation. Something like this:

    if apt-key fingerprint 0EBFCD88 | grep &quot;Key fingerprint = 9DC8 5822 9FC7 DD38 854A E2D8 8D81 803C 0EBF CD88&quot; &gt; /dev/null
    then
      # proceed
    fi

This worked fine in older versions of `apt-key`, but recent versions have two issues:

1. A different output format: I can hack around that
2. A warning:

        Warning: apt-key output should not be parsed (stdout is not a terminal)

Clearly, I can hack around this as well, just redirect `stderr` to `/dev/null`. It just made me curious:

**How do these fine folks suggest I verify my key fingerprints?** Or am I getting this fundamentally wrong by wanting to automate it, does that defeat the point? (I think not, since I still manually lifted the expected fingerprint from the website, but feel free to tell me otherwise...)

How to circumvent &quot;apt-key output should not be parsed&quot;?

I am learning to use &#39;pillow 5.0&#39; following book &#39;Automate the boring stuff with python&#39;

The info about the image object

    In [79]: audacious = auda
    In [80]: print(audacious.format, audacious.size, audacious.mode)
    PNG (1094, 960) RGBA

When I tried to convert filetype, it report error.

    In [83]: audacious.save(&#39;audacious.jpg&#39;)
    OSError: cannot write mode RGBA as JPEG
There&#39;s no such a
n error in book.

cannot write mode RGBA as JPEG

I am learning to use 'pillow 5.0' following book 'Automate the boring stuff with python'
The info about the image object
<pre><code class="hljs language-ini">In [79]: audacious = auda
In [80]: print(audacious.format, audacious.size, audacious.mode)
PNG (1094, 960) RGBA
</code></pre>
When I tried to convert filetype, it report error.
<pre><code class="hljs language-css">In [83]: audacious.save('audacious.jpg')
OSError: cannot write mode RGBA as JPEG
</code></pre>
There's no such a
n error in book.

Recently I have been working with Typescript a lot, it allows to express things like:

    interface Address {
        street: string;
        housenumber: number;
        housenumberPostfix?: string;
    }

    interface Person {
        name: string;
        adresses: Address[]
    }
    
    const person: Person = {
        name: &#39;Joe&#39;,
        adresses: [
            { street: &#39;Sesame&#39;, housenumber: 1 },
            { street: &#39;Baker&#39;, housenumber: 221, housenumberPostfix: &#39;b&#39; }
        ]
    }

Pretty concise and giving all the luxuries as type checking and code completion while coding with Persons.

How is this done in Python?

I have been looking at Mypy and ABC but did not yet succeed in finding the pythonic way to do something similar as the above (my attempts resulted in way too much boilerplate to my taste).


Python equivalent of Typescript interface

I&#39;ve just started to experiment with AWS SageMaker and would like to load data from an S3 bucket into a pandas dataframe in my SageMaker python jupyter notebook for analysis.

I could use boto to grab the data from S3, but I&#39;m wondering whether there is a more elegant method as part of the SageMaker framework to do this in my python code?

Thanks in advance for any advice.

Load S3 Data into AWS SageMaker Notebook

I use the following code when training a model in keras

    from keras.callbacks import EarlyStopping

    model = Sequential()
    model.add(Dense(100, activation=&#39;relu&#39;, input_shape = input_shape))
    model.add(Dense(1))
    
    model_2.compile(optimizer=&#39;adam&#39;, loss=&#39;mean_squared_error&#39;, metrics=[&#39;accuracy&#39;])


    model.fit(X, y, epochs=15, validation_split=0.4, callbacks=[early_stopping_monitor], verbose=False)

    model.predict(X_test)

but recently I wanted to get the best trained model saved as the data I am training on gives a lot of peaks in &quot;high val_loss vs epochs&quot; graph and I want to use the best one possible yet from the model.

Is there any method or function to help with that?

Saving best model in keras

I am using Python 3.6.1, and I have come across something very strange. I had a simple dictionary assignment typo that took me a long time to find.

    context = {}
    context[&quot;a&quot;]: 2
    print(context)

Output

    {}

What is the code `context[&quot;a&quot;]: 2` doing? It doesn&#39;t raise a `SyntaxError` when it should IMO. At first I thought it was creating a slice. However, typing `repr(context[&quot;a&quot;]: 2)` raises a `SyntaxError`. I also typed `context[&quot;a&quot;]: 2` in the console and the console didn&#39;t print anything. I thought maybe it returned `None`, but I&#39;m not so sure.

I&#39;ve also thought it could be a single line if statement, but that shouldn&#39;t be the right syntax either.

Additionally, `context[&quot;a&quot;]` should raise a `KeyError`.

I am perplexed. What is going on?

What is this odd colon behavior doing?

Is it possible in PyTorch to change the learning rate of the optimizer in the middle of training dynamically (I don&#39;t want to define a learning rate schedule beforehand)?

So let&#39;s say I have an optimizer:

    optim = torch.optim.SGD(model.parameters(), lr=0.01)

Now due to some tests which I perform during training, I realize my learning rate is too high so I want to change it to say `0.001`. There doesn&#39;t seem to be a method `optim.set_lr(0.001)` but is there some way to do this? 

PyTorch: How to change the learning rate of an optimizer at any given moment (no LR schedule)

So I&#39;m designing a few programs for editing photos in `python` using `PIL` and one of them was converting an image to greyscale (I&#39;m avoiding the use of any functions from `PIL`).

The algorithm I&#39;ve employed is simple: for each pixel (colour-depth is 24), I&#39;ve calculated the average of the `R`, `G` and `B` values and set the RGB values to this average.

My program was producing greyscale images which seemed accurate, but I was wondering if I&#39;d employed the correct algorithm, and I came across [this answer][1] to a question, where it seems that the &#39;correct&#39; algorithm is to calculate `0.299 R + 0.587 G + 0.114 B `.

I decided to compare my program to this algorithm. I generated a greyscale image using my program and another one (using the same input) from [a website online][2] (the top Google result for `&#39;image to grayscale&#39;`.

To my naked eye, it seemed that they were exactly the same, and if there was any variation, I couldn&#39;t see it. However, I decided to use [this website][3] (top Google result for `&#39;compare two images online&#39;`) to compare my greyscale images. It turned out that deep in the pixels, they had slight variations, but none which were perceivable to the human eye at a first glance (differences can be spotted, but usually only when the images are laid upon each other or switched between within milliseconds).

**My Questions (the first is the main question)**: 

 1. **Are there any disadvantages to using my &#39;rough&#39; greyscale algorithm?**
 2. **Does anyone have any input images where my greyscale algorithm would produce a visibly different image to the one that would be &#39;correct&#39; ?**
 3. **Are there any colours/RBG combinations for which my algorithm won&#39;t work as well?**

My key piece of code (if needed):

    def greyScale(pixelTuple):
        return tuple([round(sum(pixelTuple) / 3)] * 3)

The &#39;correct&#39; algorithm (which seems to heavily weight green):

    def greyScale(pixelTuple):
        return tuple([round(0.299 * pixelTuple[0] + 0.587 * pixelTuple[1] + 0.114 * pixelTuple[2])] * 3)

My input image:
[![My input image][4]][4]

The greyscale image my algorithm produces:
[![The greyscale image my algorithm produces][5]][5]

The greyscale image which is &#39;correct&#39;:
[![The greyscale image which is &#39;correct&#39;][6]][6]

When the greyscale images are compared online (highlighted red are the differences, using a fuzz of 10%):
[![When the greyscale images are compared online (highlighted red are the differences, using a fuzz of 10%)][7]][7]

Despite the variations in pixels highlighted above, the greyscale images above appear as nearly the exact same (at least, to me).

Also, regarding my first question, if anyone&#39;s interested, [this site][8] has done some analysis on different algorithms for conversions to greyscale and also has some custom algorithms.


**EDIT**:

In response to @Szulat&#39;s answer, my algorithm **actually** produces this image instead (ignore the bad cropping, the original image had three circles but I only needed the first one):

[![This is what my algorithm **actually** produces][9]][9]

In case people are wondering what the reason for converting to greyscale is (as it seems that the algorithm depends on the purpose), I&#39;m just making some simple photo editing tools in `python` so that I can have a mini-Photoshop and don&#39;t need to rely on the Internet to apply filters and effects.

**Reason for Bounty**: Different answers here are covering different things, which are all relevant and helpful. This makes it quite difficult to choose which answer to accept. I&#39;ve started a bounty because I like a few answers listed here, but also because it&#39;d be nice to have a single answer which covers everything I need for this question.

  [1]: https://stackoverflow.com/a/12201744/7908770
  [2]: https://pinetools.com/grayscale-image
  [3]: https://online-image-comparison.com/
  [4]: https://i.stack.imgur.com/7rtQk.jpg
  [5]: https://i.stack.imgur.com/FvYYq.jpg
  [6]: https://i.stack.imgur.com/jECAG.png
  [7]: https://i.stack.imgur.com/kgCMN.jpg
  [8]: http://www.tannerhelland.com/3643/grayscale-image-algorithm-vb6/
  [9]: https://i.stack.imgur.com/GIFDX.png

Problems with using a rough greyscale algorithm?

I have an image stored in a numpy array that I want to convert to `PIL.Image` in order to perform an interpolation only available with PIL.

When trying to convert it through `Image.fromarray()` it raises the following error:

&gt; TypeError: Cannot handle this data type

I have read the answers [here](https://stackoverflow.com/questions/7700193/pil-cannot-handle-this-data-type) and [here](https://stackoverflow.com/questions/51979554/typeerror-cannot-handle-the-data-type-in-pil-image) but they do not seem to help in my situation.

What I&#39;m trying to run:

    from PIL import Image

    x  # a numpy array representing an image, shape: (256, 256, 3)

    Image.fromarray(x)

PIL TypeError: Cannot handle this data type

I am moving my config from setup.py to setup.cfg and having issues setting up the `entry_points` parameter. At the moment I am using a hybrid approach which works, however, I would like to move the `entry_points` to setup.cfg.

From 

    def setup_package():
        setup(version=get_version(),
              entry_points={&#39;console_scripts&#39;:[&#39;app=my_package.app.run:cli&#39;],})

to

```cfg
[metadata]
name = my-package
description = my-package
license = unlicensed
long-description = README.md
platforms = any
classifiers =
  Programming Language :: Python

[options]
zip_safe = False
packages = my_package,  my_package.app
include_package_data = True
package_dir =
  = .
tests_require = pytest; pytest-cov

[entry_points]
console_scripts =
  my-package = my_package.app.run:cli
```

How to set up entry_points in setup.cfg

Suppose you want to constrain a type variable to implement a certain interface. You might write something like so:

    from typing import TypeVar, Callable

    T = TypeVar(&#39;T&#39;, Callable)

    class Foo(Generic[T]):
        ...

    &gt;&gt; TypeError: A single constraint is not allowed

Why is Python unhappy about this use of type constraints? [PEP 484](https://www.python.org/dev/peps/pep-0484/#generics) and the [Python source code](https://github.com/python/typing/blob/master/src/typing.py#L511) are unhelpful in this regard.

Note: in my particular case I am interested in constraining a type variable to implement an abstract base class, but the principle is the same.

Why are single type constraints disallowed in Python?

I&#39;m trying to install tensorflow onto a Mac with Python3.7. However, I&#39;m getting the error:

    $ pip3 -v install tensorflow
    ...    
        Skipping link https://files.pythonhosted.org/packages/56/7a/c6bca0fe52a94ca508731d8b139e7dbd5a36cddc64c19f422f97e5a853e8/tensorflow-1.10.0rc1-cp36-cp36m-win_amd64.whl#sha256=3ab24374888d6a13d55ce2e3cf4ba0c9cd6f824723313db5322512087525cb78 (from https://pypi.org/simple/tensorflow/); it is not compatible with this Python
      Could not find a version that satisfies the requirement tensorflow (from versions: )
    Cleaning up...
    Removed build tracker &#39;/private/var/folders/4n/9342s4wd3jv0qzwjz8rxrygr0000gp/T/pip-req-tracker-3p60r2lo&#39;

    No matching distribution found for tensorflow

From what I can gather this is happening because tensorflow doesn&#39;t yet support Python3.7. As a workaround I want to install Python3.6 alongside 3.7 and then install tensorflow to that version. However, I&#39;m new to Mac and not sure of the correct way to do this without potentially messing with the preexisting Python version. 

I&#39;ve tried using brew, but it looks like Python3 is as specific as it gets. What is the correct way to do what I&#39;m after?

Installing Python3.6 alongside Python3.7 on Mac

I tried to read https://hackernoon.com/asynchronous-python-45df84b82434.
It&#39;s about asynchronous python and I tried the code from this, but I&#39;m getting a weird Error.
The code is:
`

    import asyncio
    import aiohttp
    
    urls = [&#39;http://www.google.com&#39;, &#39;http://www.yandex.ru&#39;, &#39;http://www.python.org&#39;]
    
    async def call_url(url):
        print(&#39;Starting {}&#39;.format(url))
        response = await aiohttp.ClientSession().get(url)
        data = await response.text()
        print(&#39;{}: {} bytes: {}&#39;.format(url, len(data), data))
        return data
    
    futures = [call_url(url) for url in urls]
    
    asyncio.run(asyncio.wait(futures))


When I try to run it says:

    
    Traceback (most recent call last):
      File &quot;test.py&quot;, line 15, in &lt;module&gt;
        asyncio.run(asyncio.wait(futures))
    AttributeError: module &#39;asyncio&#39; has no attribute &#39;run&#39;
    sys:1: RuntimeWarning: coroutine &#39;call_url&#39; was never awaited
    
    

I dont have any files named ayncio and I have proof:

    &gt;&gt;&gt; asyncio
    &lt;module &#39;asyncio&#39; from &#39;/usr/lib/python3.6/asyncio/__init__.py&#39;&gt;


Python3.6 AttributeError: module &#39;asyncio&#39; has no attribute &#39;run&#39;

I&#39;ve recently upgraded from `Ubuntu 18.04` to `19.04` which has `python 3.7`. But I work on many projects using `Python 3.6`. 

Now when I try to create a `virtualenv` with `Python 36` in PyCharm, it raises:

    ModuleNotFoundError: No module named &#39;distutils.core&#39;
[![enter image description here][1]][1]

I can&#39;t figure out what to do. 

I tried to install distutils:

    milano@milano-PC:~$ sudo apt-get install python3-distutils
    Reading package lists... Done
    Building dependency tree       
    Reading state information... Done
    python3-distutils is already the newest version (3.7.3-1ubuntu1).
    0 upgraded, 0 newly installed, 0 to remove and 0 not upgraded.

But as you can see I have the newest version.

Do you know what to do?

  [1]: https://i.stack.imgur.com/kKMZC.png

Content Type	Original Author	Original Content on Stackoverflow
Question	user9062604	View Question on Stackoverflow
Solution 1 - Python	Patrick Artner	View Answer on Stackoverflow

cannot write mode RGBA as JPEG

Python Problem Overview

Python Solutions

Solution 1 - Python

How to circumvent "apt-key output should not be parsed"?

Removing broken packages in Ubuntu

Attributions