[Deep Learning Coursea] Why is my output different from he did?

Get Help
#1

Hello guys
in the ‘deep learning with tensorflow’ course at lesson ’ perceptron gate’, I copied the code he did in the video to run, why am I getting a different output from him?

image
image1288×870 152 KB

My purple line is in the middle, and the purple one he reworked in the upper right corner, wheras my code and he are the same.

link to video:https://youtu.be/grb4LlTg1RE
link to course: https://www.codecademy.com/paths/build-deep-learning-models-with-tensorflow/tracks/dlsp-foundations-deep-learning-and-perceptrons/modules/perceptron/projects/perceptron-logic-gates

#2

This post was flagged by the community and is temporarily hidden.

#3

link to the video he did

#4

This post was flagged by the community and is temporarily hidden.

#5

There’s some discussion on it here that might be of interest to you-

#6

please help tell me why

#7

I believe the reason is statistics. In the distributions I made you can see the long tailed distribution on both parameters when 4 data points are used to fit the model versus the very tight distribution that is produced when using 1000 data points. The model learns the rules “better” the more training data it has so it can consistently produce the same model.

#8

i dont understand, i am not good at statistics, i I forgot my statistical knowledge, but why he had the true result ?
i have a false result

#9

1000 data points are only for testing, not for training, you got it wrong, there are only 4 data points for training.

#10

I did 4 data points for training and 10000 for training running each 1000 times. Here’s the code:

import numpy as np                                                        
from sklearn.linear_model import Perceptron                               
from itertools import product   
from random import randint
import matplotlib.pyplot as plt
from collections import Counter

validation = [[0,0], [1,1],[0.5,0.5]]

def generate(m):
   '''Generates a data and label set of size m'''

   d,l = [], [] 
   for i in range(m): 
       x,y = randint(0,1), randint(0,1) 
       d.append([x,y]) 
       l.append(x and y) 
   return d,l 

def large(n, m):
   '''Takes number of simulations, n, and number of data
   Points per data set, m, as input. Returns two Counter
   Objects: coefficients and decision function results'''

   coef_set = [] 
   dv_set = [] 
   for i in range(n): 
       d,l = [],[] 
       while len(set(l)) < 2: 
           d,l = generate(m) 
       clf = Perceptron(max_iter=40) 
       clf.fit(d,l) 
       coef_set.append(tuple(clf.coef_.tolist()[0])) 
       dv_set.append(tuple(clf.decision_function(validation).tolist())) 
       print('\r',str(round((i+1)/n *100, 2))+'%', end='', flush=True) 
   return Counter(coef_set), Counter(dv_set)

def graph(x,color='blue', title=''):
    '''Takes Counter object, title and color and generates
    distribution plot'''

    labels,values = zip(*x.items()) 
    i = np.arange(len(labels)) 
    values = tuple([v/sum(values) for v in values]) 
    width = 1 
    plt.figure(figsize=(12,5)) 
    plt.bar(i,values,width,color=color, alpha=0.5) 
    plt.xticks(i,labels) 
    plt.title(title) 
    plt.show()

# example usage
c, d = large(1000,4)


graph(c,'red',
    'Coefficent Distribution: 1000 iterations with 4 data points'
)

graph(d,'blue',
    'Decision Function Distribution: 1000 iterations with 4 data points'
)
1 Like