Attention is not enough! Mathematical Representation of Words -embedding vectors

Figure3.py

from manim import *
from manim.utils.unit import Percent, Pixels
from colour import Color

import gensim
from gensim.models import Word2Vec
import numpy as np

import Util_Functions
from Util_Functions import *

config.frame_width = 32
config.frame_height = 18

HIGH = True
if HIGH:
    print("Its high")
    config.pixel_height = 2160
    config.pixel_width = 3840
else:
    print("Its low")
    config.pixel_height = 1080
    config.pixel_width = 1920

data = [['Attention', 'is', 'not', 'enough'], ['Towards', 'Data', 'Science'], ['It', 'was', 'a', 'really', 'good', 'article']]

sentences_Embedded = construct_Word_Embeddings(data, 6)

sentence_split = data[0]
sentence_embed = np.round(sentences_Embedded[0],2)



class Figure3(Scene):
    def construct(self):

        ################################################################################
        #                    INTRO ANIMATION SAME ACROSS ALL VIDEOS                    #
        ################################################################################
        def myAnimation(wordsForIntro: str):
            fontHeight = config.frame_height//8
            fontColor = WHITE
            timePerChar = 0.1
            C = MathTex(r"\mathbb{C}", color = fontColor).scale(config.frame_height//3)
            self.play(Broadcast(C), run_time=1)
            self.add(C)

            # Building text objects of individual characters.
            wordsMath = VGroup()
            for word in wordsForIntro:
                charTex = VGroup()
                for i,ch in enumerate(word):
                    chTex = MathTex("\mathbb{%s}"%ch, color = fontColor).scale(fontHeight)
                    if i != 0:
                        chTex.next_to(charTex[-1], RIGHT, buff=0.05).align_to(C, DOWN)
                    else:
                        chTex.next_to(C, RIGHT, buff=0.05).align_to(C, DOWN)
                    charTex.add(chTex)
                wordsMath.add(charTex)
            
            # Succesion or AnimationGroup--- Both are messed up ----HENCE INEFFECIENT ANIMATION
            for wInd in range(len(wordsMath)):
                for chInd in range(len(wordsMath[wInd])):
                    self.play(Write(wordsMath[wInd][chInd], run_time = timePerChar))
                self.wait(0.5)
                for chInd in reversed(range(len(wordsMath[wInd]))):
                    self.play(Unwrite(wordsMath[wInd][chInd], run_time = timePerChar))

            self.play(Circumscribe(C, color=MAROON_E, fade_out=False, time_width=2, shape=Circle, buff=1, stroke_width=config.frame_height//3, run_time=1.5))
            self.play(ShrinkToCenter(C, run_time=0.25))
            self.wait(0.5)
        # myAnimation(wordsForIntro= ['OMPLEX', 'ONCEPTS', 'OMPREHENSIBLE'])

        ################################################################################
        #                    MAIN CODE OF THE FIGURE STARTS HERE                       #
        #                   NO FUNCTION OR ANYTHING OF THAT SORT                       #
        ################################################################################

        # Building Mobjects of Word Vectors and arranging them horizontally
        Word_Vectors = VGroup( 
                *[ Matrix(word.reshape(-1,1)).set_column_colors([GREEN_A,GREEN_E]) for word in sentence_embed]
            )
        Word_Vectors.arrange(RIGHT, buff=1)

        # Building Text Mobjects of words corresponding to Word Vectors (May replace it in future if I find how to caption Mobjects)
        Words = VGroup(
                *[Text(word,
                       slant = ITALIC,
                       color = BLUE_C,
                       font_size = 64) for word in sentence_split]
            )
        
        for i in range(len(Word_Vectors)):
            Words[i].next_to(Word_Vectors[i], DOWN*2)

        # Caption = Text("Embeddings are high dimensional-vectors that represent words \n and its semantics mathematically since models cannot interpret words directly.",
        #                slant = ITALIC,
        #                line_spacing=1,
        #                font_size = 32).next_to(Word_Vectors, UP*2)

        # Creating and triggering the animation
        self.play(Create(Word_Vectors, run_time = 4))
        self.play(
            AnimationGroup(
                *[ DrawBorderThenFill(word) for word in Words],
                lag_ratio = 1
            ), run_time = 4
        )

        self.play(
            AnimationGroup(
                *[ ApplyWave(word, ripples=4) for word in Words],
                lag_ratio = 1
            ), run_time = 4
        )

        self.wait(8)