Epoch in Neural Network|neural network example step by step |Neural network end to end example data
Вставка
- Опубліковано 15 вер 2024
- Epoch in Neural Network|neural network example step by step |Neural network end to end example data
#NeuralNetworkEpoch #NeuralnetworkbackPropagation #UnoldDataScience
Hello All,
Welcome to Unfold data science, this is Aman and I am a data scientist. In this video I explain every step of neural network with data. I explain how forward propagation and backward propagation works. I explain how weights are adjusted using gradient descent.Below questions are answered this video:
1. How neural network work?
2. Neural Network example step by step
3. How forward propagation works in neural network
4. How backward propagation works in neural network
5. How neural network adjusts weight
6. What is a neural network epoch?
About Unfold Data science: This channel is to help people understand basics of data science through simple examples in easy way. Anybody without having prior knowledge of computer programming or statistics or machine learning and artificial intelligence can get an understanding of data science at high level through this channel. The videos uploaded will not be very technical in nature and hence it can be easily grasped by viewers from different background as well.
Join Facebook group :
www.facebook.c...
Follow on medium : / amanrai77
Follow on quora: www.quora.com/...
Follow on twitter : @unfoldds
Get connected on LinkedIn : / aman-kumar-b4881440
Follow on Instagram : unfolddatascience
Watch Introduction to Data Science full playlist here : • Data Science In 15 Min...
Watch python for data science playlist here:
• Python Basics For Data...
Watch statistics and mathematics playlist here :
• Measures of Central Te...
Watch End to End Implementation of a simple machine learning model in Python here:
• How Does Machine Learn...
Learn Ensemble Model, Bagging and Boosting here:
• Introduction to Ensemb...
Access all my codes here:
drive.google.c...
Have question for me? Ask me here : docs.google.co...
My Music: www.bensound.c...
You are a hero! thank you for simplifying such very complicated concepts like this!
just 5k views?! This video is Gold...even the best youtube channels on NN can't explain you what this SIR here has taught us....the level of simplicity is just too high.....tysm for sharing your knowledge.....this explaination of one entire epoch is nowhere else.
Thanks Anuj for motivating me through your comments.
OMG!!!!!!............I can't believe I watched this video for free!
Thanks a lot.
Simplest explanation I've found on how the epoch works.... Muchas gracias!
I m a data scientist with 10+ years experience, but I must say this is by far the best explanation possible. I am eagerly waiting for many more videos from you. Keep us enlightened 🙂👍👌.
I searched various videos related deep learning in youtube. I didn’t get correct explanation from them..I have come across your video..wow your explanation excellent and easy to understand..you are best👍
Hi Aman, this is the best video I have come across to explain the maths... simple brilliant!!!!!
Thanks Sam. I am happy you could understand.
Thanks so much from Ghana, I have followed from the beginning to this point and for the first time I am able to comprehend what basic of neural networks, and the maths behind it. Please I will highly appreciate it if you can do a video for the neural network design and choosing the right architecture given a real world problem, before we start implementing in codes. Thank you.
You are so welcome
God bless you! I wish I could subscribe a thousand times to your channel. Best content ever on DL
Your comments are precious to me.
@@UnfoldDataScience You have the gift of knowing how to simplify things. By the way, do you have a plan on teaching Reinforcement Learning in the future?
At 6:51, the outputs 0.7513 from O1 and 0.7729 from O2 are "after passing their respective inputs to sigmoid activation function". Missed to write in whiteboard.
tanks a lot for all clarification . here only i was having doubt.
Made simple and clear. Thank you.
Thanks a lot :)
Thank you! The video was very simple yet helpful!
Glad it helped!
Thanks for a very clear explanation on what epoch is.
Welcome Mathew. Your comments are precious for me
Hi Aman, your explanation is very very clear and anybody can understand easily. Thanks
Nice and easy explanation. @Unfold Data Science can you suggest good book which has easy way of understanding and all the above mathematics
Simple and amazing!!! Congratulations !!!!
Thanks Kese
First of all, great explanation - very easy to understand, thanks much! When I was pondering further at the end part when we calculated the error, I got a question. What if we had 8 input neurons, 12 in the first layer, 8 in the second hidden layer, and one neuron in the output layer in case of a regression problem (you have a video on similar structure to demo capability of keras). Let's say we took a batch of 10 rows - first iteration. For each of 10 rows, there will be one target value in our data. Question is how will we end up with the error value at the end on the output neuron? Does it happen the way that after the second hidden layer, we get a weight and a bias as input to the output layer - the values then applied to all the 10 rows to get individual prediction. (How does it get back original values at this stage?) Each predicted value will be compared with the actual value, and then for all the 10 rows the sum squared will be averaged out? which will be the final error coming out from the output layer. Then this error will be used to find gradient which will be applied to optimize weights at individual neurons in hidden layer? Hope, my doubt is understood.
Thanks aman.....great explanation 🙏
Welcome Abirami :)
Excellent bro. Nice and indepth video. Kudos to you. One doubt - at 11th min the Gradient for W5 calculated was 0.08. But when calculating W5 new value you made it 0.8. Ideally should it not be W5 newvalue = (.40) - (0.1 x .08) = .40 x 0.008 = 0.392 ?. Please clarify.
Let me check Indrajith. Thanks for the feedback.
you said it right
Superb sir thank you but please take feed back and implement, new people difficult to get concept
I will try my best
very good teaching !!!!!!
Thanks you very much for a great conceptual explanation.
You are welcome Brij Kishor.
simply helpful
Thank you Ravi.
Amazing job. Keep up the good work. Understood completely.
Thanks Akshay. Happy learning. tc
underrated channel
Really good explanation.....
Thanks Deepak :)
Thanks for the video, nice explanation
Glad it was helpful Bhabesh.
Pls give the calculations in details... Specifically partial derivatives of chain rules applied in your final calculations in minimization.
Really great explanation, and helpful
Glad it was helpful Debjyoti.
Hi Aman,
Wonderful explanation. BTW can you please let me know what is the meaning of 'num_epcoh' value we give in training?
Hi Rajavel, it tells to python how many epochs we want in the model, this is just a parameter we can control actually.
Hi, during parameter optimization example, for the first part, while using the partial derivate of E_total w.r.t Out_o1, why you multiplied by (-1) after which you got a final value of 0.7413? It is between 10:08 and 10:40 in the video. Can you please explain?
Wow! You are really good
Thank you Nargis 😊
Wounderfull explanation sir,but how accuracy is calculated for one epoch?
I know I'm late but if it works of something, the accuracy it's not the thing that is calculated, the objective of every epoch is to minimize the error till the point it reach 0 or pretty close(as close the error to 0, the more accurate is the model), the error is minimized when you update the weights in every epoch , so the algorithm it comes to the end, after completing the epoch you define, or the error cost reach 0, thing that is very unlike to happen
Nice explaination
Keep watching Uday.
Thanks from Lebanon
Welcome Youssef :)
Hi Aman , Wonderful work :)
Quick question here , this whole error that you have calculated is for one observation or this is for the entire dataset that we have fed into the network ?
Does Output 1 indicates the output of the first row and so on ? It means we have 50 observations then , we would have 50 outputs at the end ? Please respond . Thanks
great explanation
Glad you think so Sandipan.
Well explained 👍
Thank you 🙂
Thank you for the video. Can you explain the difference between 1 epoch from another epoch? Are the weights adjusted at each iteration or at each epoch?
If its a full batch processing then 1 iteration = 1 epoch.
Weights are adjusted in various iteratio
'Are the weights adjusted at each iteration or at each epoch?
'-weights will be adjusted after each batch. Am I right, @unfold data science?
'difference between 1 epoch from another epoch'. the next epoch will do the same stuff, the difference is, the next epoch is based on the updated weights which were adjusted earlier on. Am i correct, @Unfold Data Science?
great explanation!
Glad it was helpful!
Great explanation aman. Just one thing, the gradient you calculated was 0.08, but you wrote 0.8
I will check, Thanks for pointing out Akash.
how can I calculate the gradient from the hidden layer to the input layer , is it E/outH1 x outH1/l1 x l1/w1 ?? because I didn't find anyone explaining that
Can you please make one video on back propagation
Go to playlist, find neural network playlist, you will find all videos on this topic there.
Hi Aman,
Thank you so much for explaining this concept.
But I have few doubts like
#.) what is the essence or say need of activation function? Why do we need them? What will happen if we don't use them, and for each node, what goes as input will come out as output?
Without activation function its a linear regression model.
Very nice Aman
Thanks a lot :)
Hi Aman,
So Basically What I understood from your video is that an epoch is the number of iterations that an algorithm has to work to optimize the weights and biases to minimize the error function for each training dataset.
Please correct me if i am wrong.
Epoch is one cycle completed for entire training data.
suppose you have 100 rows in training data and you divide into 10 batches.
One iteration means, covering one batch that is 10 records
One epoch means, covering 100 records that is10 iterations.
Thanks for explanation it was wonderful,
what can we get if increase and decrease the epoch?
Hi Mohammed, model performance might change with more shift.
Hi Aman,I am using Rasa during model training my EPOCHS are taking long time to load should I use NPU in the server for faster loading...currently using GPU which is faster than CPU
Here you updated only one parameter. One Epoch = all the weights and biases are updated once. Right?
Yes Soumya, for demo purpose just one parameter I showed but 1 epoch means all the weight and bias get updated. You are absolutely right.
hello there!, as per knowledge each epoch has a training and validating phase, can you explain the validation phase??
at last , gradient is 0.08 then why are we taking 0.8 . also how is the step size 0.1 , can it be also 0.01? please help
I have a question.i stake cardano in a wallet that give me 5percent and it epoch my crypto and every 5 day it begin its approach process,how it work and how much cardano i give fron out put if my input is 69.
thanks
Hi aman sir,
Can you post a video on dropout layers for regularization in Deep neural network?..it will be very useful for us i guess..
Thanks in advance. @Unfold Data Science
It's not at all with number that someone would understand a concept. You must do the analogy with speed and acceleration to explain the gradient descent
Good suggestion, I explain GD here:
ua-cam.com/video/gzrQvzYEvYc/v-deo.html
Do values in input layer (x1, x2, x3...) represent features in single data row.. or no. of rows of data?
This is my doubt. How multiple row of taken place
many rows
Sir in feed forward neural networks how will the parameter updation take place since there is no feedback loop??
So if give the batch size for one epoch as 10, does this mean that in the total_error, we will sum up all the errors?
Yes for that batch.
@@UnfoldDataScience great thanks for confirming that
Aman bai, please explain difference between adam optimizer and sgd please...
Thanks for the feedback Santosh. Will create Video on suggested topic. Happy Learning. Tc
(1) What happens if you are training on a batch of records? Does that mean, once all the rows in the batch is completed, that is equivalent to one EPOCH. (2) If it is not being batched, and training is occurring on a row-by-row basis, then a full EPOCH is 1 row being traininged?
Hi Karun,
Good question.
If you are training on batch or non-batch, in any case, one epoch is when all the records complete forward and backward propagation once.
Could you explain a bit why doing more epochs could lead to overfitting ?
Yes, because of many iterations, the parameters may get tuned in a way that might lead to fitting problem.
Hi Sir, do you have any video on conjugate gradient training method, if not could you please make the one.
Will upload soon
One doubt i have, what we see in epoch accuracy , that accuracy is before changing weight or after changing weight, because epoch first calculate total error....could not get the given accuracy is before changed or after
epoch 10 accuracy is after adjusting the weight in epoch 9. Just an example.
can help me i wanna prediction by ANN in Matlab?????plz
Can you tell me how to simulate random binary noise for random inputs and targets?
Not able to recollect now. Will do some research and let u know.
Hi: What is the relationship between epoch and accuracy????
Good question Shadab,
More epoch lead to better model with high accuracy as model has more learning opportunity.
May not be always true though. Normally it happens.ok
sir can you please guide me what is the right way to begin with machine learning , neural networking if I am now in 12th
Please start learning statistics and python coding.
@@UnfoldDataScience thankyou sir for replying and sir i already know to code in python
who is defining actual value in error? at 7:19
Hi Shantanu, It comes from training data.
what if you dont have bias do you just do the sum of xiwi then?
Correct, Bias is zero hence only xiwi.