Installing under Anaconda

[sck-at-ucy]

I have been unable to get Manim_ml to properly install under Anaconda distribution. Any advise?

[helblazer811]

You ran pip install manim_ml and it didn't work?

[sck-at-ucy]

It seemed to install OK. But when I try to run one of the examples, I get this error

---

ImportError Traceback (most recent call last)
Cell In[2], line 3
1 from manim import *
----> 3 from manim_ml.neural_network import Convolutional2DLayer, FeedForwardLayer, NeuralNetwork
5 # This changes the resolution of our rendered videos
6 config.pixel_height = 700

ImportError: cannot import name 'Convolutional2DLayer' from 'manim_ml.neural_network' (/usr/local/anaconda3/envs/my-manim-environment/lib/python3.8/site-packages/manim_ml/neural_network/init.py)

[sck-at-ucy]

Wonder if it is a path issue?

[helblazer811]

Hmmm. I tried to simplify imports by having everything importable from manim_ml.neural_network. There may be a bug in how that is done.

Can you try and do from manim_ml.neural_network.layers.convolutional_2d import Convolutional2DLayer?

Does that fix the issue?

[helblazer811]

Also can you specify which version of manim_ml you have installed? You can do this by running pip show manim_ml.

[sck-at-ucy]

I made the change and was able to import

[sck-at-ucy]

Name: manim-ml
Version: 0.0.13
Summary: Machine Learning Animations in python using Manim.
Home-page: https://github.com/helblazer811/ManimMachineLearning
Author: Alec Helbling
Author-email:
License:
Location: /usr/local/anaconda3/envs/tf/lib/python3.10/site-packages
Requires:
Required-by:
Note: you may need to restart the kernel to use updated packages.

[sck-at-ucy]

I should mention that I am trying all these inside a Jupyter notebook

[helblazer811]

I made the change and was able to import

Awesome. I'll look into the importing issue. Thanks for the help!

[helblazer811]

I should mention that I am trying all these inside a Jupyter notebook

I have not tested this at all in Jupyter notebook environments. So it may not work there. However, it looks like the core ManimCommunity library did work to support Jupyter.

https://docs.manim.community/en/stable/installation/jupyter.html

[sck-at-ucy]

Yes, I am able to run all the standard examples for ManimCommunity from inside Jupyter notebooks.

Thank you, let me know if you want me to do any further tests.

[helblazer811]

Amazing! Feel free to go through more of the examples in the examples/ directory. I didn't expect all of the attention so I haven't tested everything as thoroughly as I should have.

[sck-at-ucy]

OK, I sort of got it working in Jupyter. I think something is broken with a path (at least on my installation) but the two things I did that got it work are:

1. Changed this part in example code because it was complaining about Neural_Network()

# The code for generating our scene goes here
    def construct(self):
        # Make the neural network
        nn = NeuralNetwork([.     # <------------------------------------Changed here!
                Convolutional2DLayer(1, 7, 3, filter_spacing=0.32),
                Convolutional2DLayer(3, 5, 3, filter_spacing=0.32),
                Convolutional2DLayer(5, 3, 3, filter_spacing=0.18),
                FeedForwardLayer(3),
                FeedForwardLayer(3),
            ],
            layer_spacing=0.25,
        )

2. Before running the example, in the previous code cell of Jupyter I simply run

# %load '/usr/local/anaconda3/envs/my-manim-environment/lib/python3.8/site-packages/manim_ml/neural_network/neural_network.py'
![jupyter_manim](https://user-images.githubusercontent.com/5185237/215352183-d4112a05-0057-41a9-8f45-79ccaba0d068.png)

The second step tells me that somehow the pip installation probably did not properly set the path.

[helblazer811]

In the code, what exactly did you change?

I am not sure if you issue is specific to jupyter. I wonder if you could copy the example code at the top of the Readme into a file called example.py and try running manim -pql example.py in your command line inside of the same conda environment.

[sck-at-ucy]

This is the cell that runs successful (after loading manually the lib in the previous cell)

%%manim -pql -v WARNING BasicScene

#from manim_ml.neural_network import neural_network

#from manim_ml.neural_network.layers.convolutional_2d import Convolutional2DLayer

#from manim_ml.neural_network.layers.feed_forward import FeedForwardLayer


# This changes the resolution of our rendered videos
config.pixel_height = 700
config.pixel_width = 1900
config.frame_height = 7.0
config.frame_width = 7.0

# Here we define our basic scene
class BasicScene(ThreeDScene):

    # The code for generating our scene goes here
    def construct(self):
        # Make the neural network
        nn = NeuralNetwork([        #<------------------------------ Changed the name from Neural_Network
                Convolutional2DLayer(1, 7, 3, filter_spacing=0.32),
                Convolutional2DLayer(3, 5, 3, filter_spacing=0.32),
                Convolutional2DLayer(5, 3, 3, filter_spacing=0.18),
                FeedForwardLayer(3),
                FeedForwardLayer(3),
            ],
            layer_spacing=0.25,
        )
        # Center the neural network
        nn.move_to(ORIGIN)
        self.add(nn)
        # Make a forward pass animation
        forward_pass = nn.make_forward_pass_animation()
        # Play animation
        self.play(forward_pass)

[helblazer811]

So the name of the object is NeuralNetwork not Neural_Network. I looked through the readme and it doesn't look like that was accidentally written anywhere.

What did you do to manually load the library in the previous cell?

Again, thanks so much for all of the help!

[sck-at-ucy]

Just run this command. It loads the file and then I execute the cell.

%load '/usr/local/anaconda3/envs/my-manim-environment/lib/python3.8/site-packages/manim_ml/neural_network/neural_network.py'
![jupyter_manim](https://user-images.githubusercontent.com/5185237/215352183-d4112a05-0057-41a9-8f45-79ccaba0d068.png)

[helblazer811]

Ok. I am unsure if this is a jupyter-specific issue. I will try testing it out in a jupyter notebook.

[sck-at-ucy]

Great, thank you!

[helblazer811]

I think I just forgot to add the manim_ml/neural_network/__init__.py file which defined all of these import shortcuts in the most recent pypi package. It should be good now if you install again.

[sck-at-ucy]

It solved half of the problem in the sense that now I can use

from manim_ml.neural_network import neural_network, Convolutional2DLayer, FeedForwardLayer

to import the layers. However, I am still not able to avoid having to load manually the library, so the path issue continues.

[helblazer811]

Hmmm. Is it possible that you are not running the Jupyter notebook with the conda environment's kernel?

[sck-at-ucy]

I take it back. I restarted the kernel and go it to work like this

First cell:

from manim import *

Second cell:

%%manim -pql -v WARNING BasicScene
from manim import *

from manim_ml.neural_network import Convolutional2DLayer, FeedForwardLayer, NeuralNetwork

# This changes the resolution of our rendered videos
config.pixel_height = 700
config.pixel_width = 1900
config.frame_height = 7.0
config.frame_width = 7.0

# Here we define our basic scene
class BasicScene(ThreeDScene):

    # The code for generating our scene goes here
    def construct(self):
        # Make the neural network
        nn = NeuralNetwork([
                Convolutional2DLayer(1, 7, 3, filter_spacing=0.32),
                Convolutional2DLayer(3, 5, 3, filter_spacing=0.32),
                Convolutional2DLayer(5, 3, 3, filter_spacing=0.18),
                FeedForwardLayer(3),
                FeedForwardLayer(3),
            ],
            layer_spacing=0.25,
        )
        # Center the neural network
        nn.move_to(ORIGIN)
        self.add(nn)
        # Make a forward pass animation
        forward_pass = nn.make_forward_pass_animation()
        # Play animation
        self.play(forward_pass)

[sck-at-ucy]

By the way, I have a suggestion for future additions. It would be nice if it could include additional type of layers as used in Keras, such as Max_Pooling2D, Concatenate and Up_Sampling. That would be super!

[helblazer811]

I take it back. I restarted the kernel and go it to work like this

First cell:

from manim import *

Second cell:

%%manim -pql -v WARNING BasicScene
from manim import *

from manim_ml.neural_network import Convolutional2DLayer, FeedForwardLayer, NeuralNetwork

# This changes the resolution of our rendered videos
config.pixel_height = 700
config.pixel_width = 1900
config.frame_height = 7.0
config.frame_width = 7.0

# Here we define our basic scene
class BasicScene(ThreeDScene):

    # The code for generating our scene goes here
    def construct(self):
        # Make the neural network
        nn = NeuralNetwork([
                Convolutional2DLayer(1, 7, 3, filter_spacing=0.32),
                Convolutional2DLayer(3, 5, 3, filter_spacing=0.32),
                Convolutional2DLayer(5, 3, 3, filter_spacing=0.18),
                FeedForwardLayer(3),
                FeedForwardLayer(3),
            ],
            layer_spacing=0.25,
        )
        # Center the neural network
        nn.move_to(ORIGIN)
        self.add(nn)
        # Make a forward pass animation
        forward_pass = nn.make_forward_pass_animation()
        # Play animation
        self.play(forward_pass)

Amazing!

[helblazer811]

By the way, I have a suggestion for future additions. It would be nice if it could include the additional type of layers as used in Keras, such as Max_Pooling2D, Concatenate and Up_Sampling. That would be super!

Great ideas. I already have implemented Max_Pooling2D. You can check it out in one of the readme examples.

https://github.com/helblazer811/ManimML/blob/main/Readme.md#:~:text=Max-,Pooling,-A%20common%20operation

[sck-at-ucy]

Oh, that's becoming really nice!

[sck-at-ucy]

By the way, is padding something to be accounted for?

[helblazer811]

That’s another thing I’m planning to add soon. The feature I’m really excited about is haven’t an effect where the camera zooms in on the active layers and moves forward with the forward pass. That way I can visualize super deep architectures.

…

On Mon, Jan 30, 2023 at 2:17 PM sck-at-ucy ***@***.***> wrote: By the way, is padding something to be accounted for? — Reply to this email directly, view it on GitHub <#25 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADMGLRSYNTYMUGWCX2PRRHLWVAHUZANCNFSM6AAAAAAUKMGASM> . You are receiving this because you commented.Message ID: ***@***.***>

[sck-at-ucy]

Perfect

[helblazer811]

I added padding. For some reason I can't upload the video, but you can see it here: https://twitter.com/alec_helbling/status/1620641566024171520?s=20.

[sck-at-ucy]

It would be amazing if eventually you get it to read in the model summary files generated by Keras and then do the animation. Should’t be that hard as you already have most of the hard work.