 Do a quick stream just rotate my camera because it's easier. So I have this quick example of Quick example of a neural network training So I have this as I saw it That we are trying to predict It's shape. So for example, if it was the stock market or something we could have Try and predict it where it's gonna go next lots of fun You can control your parameters from here. So the training set size currently there was I think 18 a Data points by default you can increase it say to 30 the number of hidden layers First hidden layer second hidden layer You can also change the noise level. So say let's keep it at point five Let's see how the neural net is learning. That's essentially its performance in Green the error trend It's getting to point Point two no point one and that's the prediction result So if we add say more It is See if it does any better then we could also check how it's performing in noise. It's all Running in the java script So when deployed on the website that will be running in your browser There's a couple things That we need to fix before publishing this So currently the Styling for this page is not great Well, because it stretches the text and everything else doesn't really fit The page no it does but it's Yeah, it's still a bit funny. Yeah, you can keep changing parameters on the fly. Yes, we expect the Lower Performance with low with less data points in the training set. Yeah, so it's essentially Not being able to converge Takes some time as well Yeah, we really need a bot to if fix this Or us Page Breaks music breaks. Do you like this music or not? Let me know. I don't have a CEO or anything talking in my ears. Just be monitoring the sound from another device Just to make sure I'm using the correct Microphone Okay, so how do we improve this thing? First of all when we use GPT for we can't code we have a GitHub co-pilot there as well commented out the Child script instead of this CSS Okay, now it's better And the CSS does something to the charts Don't Charts once Over there. Hey, we'll fix it later GPT as well pop in the HTML and JavaScript now question is What in the HTML controllers is not currently being used in JavaScript as several controllers for the neural network configuration Which values Used so this is will be the activation function a Optimize A box batch size. We have a activation function option value sigmoid done Optimizer Adam CGT other grad those are standard options and The loss function is being concluded by a mean square there mean absolute error or mean squared logarithmic error This is not currently being Implemented We have to change our JavaScript Have a train model function So currently you're looking me at the training size a hidden layer size and a second layer size activation optimizer loss function a box the number of epochs and batch size a This constant model will take the activation as well We have a model Compile that will take the optimizer and the loss function and What else what else? compile Get the training that Can remain the same That they are a trend chart Yeah, what about they're a trend chart Yeah, we need to regenerate the Can we regenerate the whole train model a sync function? currently epochs and batch size a Variables are not being used see if it can do that So obviously we're using a GPT full disclosure It's right rewriting this Train model shall we split it into train model and update charts and is the separate functions? I think it's actually doing it already. It's more a to smart dynamic training configuration now using epochs and batch size correctly controlling the training loop and the number of samples per gradient update respectively Doing a chart management as a separate thing. Okay, shall we a Comment all function out this girl the team there show all the functions are Used a Okay, that's not cool box must be a much large number With the training stuff training stuff is not visible Good thing about 24 is that we Do this And they are couple of problems a First of all the training data is not visible on the top chart Many other issues Any other issues the batch size Why is this not updating the train that should be visible in the top chart? Data drinks have been correctly plotted We have a fox a learning rate Can we fix the Original thresholds and default values in the html so the whole thing works better Make sure that the training Data is actually visible on the top chart Be nice to fix the following JavaScript code as well Can it's to html adjustments first? so the learning Learning rate and the minimum value can be reduced Maximum value point three Textual weight value point oh one a number of a box needs to be increase a max of one thousand Minimum then Let's go for a value of a 100 a batch size go between 10 and 120 a default in JavaScript train model might be a problem with that Okay, we have three set Shards have the Training loop training loop this goes Oops don't get it Don't get it. You know we have the training data It's not quite a sinusoid is it a we won't prompt just Sometimes it just needs to correct the data and we will see if it It makes a difference the training that blue line is now visible. So that's great the training and prediction lines are Not overlapping much much indicated model predictions are not aligning with the training data This could be several reason. Well, also training data Should be a sinusoid That's a bit odd that it isn't Also any idea why the training data is not the sinusoid How to fix the code just be a sinusoid before a Get training that the function Incorrect data generation We are regenerating the get training data have this size as an import I can really see The difference was train model function I'm not plotting it correctly. Are we so we prepare the training that the plot it something wrong Every time you're asking it something it's generating a new function Which is not great Okay, we also need to Generate of that Visualize chances not to find what? If a kid have copart can fix this quickly Okay, I'll be finishing up in a sec. I didn't know if there's any questions about binarykills.com If you haven't checked it Already, please go do so. He will be supporting the project this way I'll say bye-bye You