Time: 2025-07-11 11:28:02 Source: Codora.ai Author: rust Reading: 636 times

used right I have science vs conspiracydiscussedmsse mean squared error before sobasically we just measuring thedeviation over here right msse isnothing but your deviation from youractual output all right that's exactlywhat we're doing here so after you'vecomputed your error the next step isobviously to update your weight and yourbias so we have something known as theoptimizers they basically take care ofall the necessary computations that areneeded to adapt the Network's weight andbias variables during the training phaseright that's exactly what's happeningover here now the main function of thisOptimizer is that it invokes somethingknown as the gradient now if you allremember we discussed gradient before itbasically indicates the direction inwhich the weights and the bias have tobe changed during the training in orderto minimize the network cost function orthe network error so you need to figureout whether you need to increase theweight and the bias in order to decreasethe error or is it the other way aroundright you need to understand therelationship between your error and yourweight variable right that's exactlywhat the optimizer does it invokes thegradient which will give you thedirection in which the weights and thebias have to be changed right so nowthat you know what an Optimizer does inour module we'll be using somethingknown as the adom optimizer this is oneof the current default optimizers indeep learning Adam basically stands foradaptive moment estimation and it can beconsidered as a combination between verytwo popular optimizers called adagradand RMS prop now let's not get into thedepth of the optimizers the main agendahere is for you to understand the logicbehind deep learning we don't have to gointo the functions and all these arepredefined functions which tensor FLtakes care of next we have somethingknown as initializers now initializerare used to initialize the networkvariables before training we alreadydiscussed this before now I Define theinitializer here again I'd already doneit earlier in this session rightinitializers are already defined so Ijust removed that line of code Next Stepwould be fitting the neural network soafter we've defined the placeholders thevariables variables which are basicallyweights and bias the initializers thecost functions and the optimizers of thenetwork the the model has to be trainednow this is usually done by using themini batch Training Method because wehave a very huge data set right so it'salways best to use the mini batchTraining Method now what happens duringmini batch training is random datasamples of any bat size are drawn fromthe training data and they are fed intothe network so the training data setgets divided into n divided by yourbatch size batches that are sequentiallyfed into the network so one after theother each of these batches will be fedinto the network at this point theplaceholders which are your X and Y theycome into play they store the input andthe target data and present them to thenetwork as inputs and targets that's themain functionality of placeholders whatthey do is they store the input and thetarget data and they provide this to thenetwork as inputs and targets that'sexactly what your placeholders do solet's say that a sampled data batch of xright now this data batch flows throughthe network until it reaches the outputlayer there the tens of flow comparesthe models predictions against theactual observed targets which is storedin y if you all remember we stored ouractual observed Targets in y after thistensor flow will conduct something knownas optimization step and it'll updatethe Network's parameters like the weightof the network and the bias so afterhaving updated your weight and the biasthe next batch is sampled and theprocess gets repeated so this procedurewill continue until all the batches havepresented to the network and one fullsweep over all batches is known as anEpoch so I've defined this entire thingover here so we are going to go through10 epochs meaning that all the batchesare going to go through training meaningyou're going to input each batch that isX and it will flow through the networkuntil it reaches the output layer therewhat happens is tens oflow will compareyour predictions that is basically whatyour model predicted against the actualobserved targets which is stored in yafter this tensorflow will performoptimization wherein it'll update thenetwork parameters like your weight andyour bias after you update the weightand the bias the next batch will getsampled and the process will keeprepeating right this happens until allthe batches are implemented in thenetwork so what I just told you was oneEpoch we're going to repeat this 10times so our batch size is 256 meaningthat we have 256 batches so here we'regoing to assign X and Y what I justspoke to you about the mini batchtraining starts over here so basicallyyour first batch will start flowingthrough the network until it reaches theoutput layer after this tensor flow Wecompare your models prediction this iswhere predictions happen it'll compareyour models prediction to the actualobserved targets which is stored in ythen tensorflow will start doingoptimization and it'll update thenetwork parameters like your weight andyour bias so after you update the weightand the biases the next batch will getinput into the network and this processwill keep repeating this process willrepeat 10 times because we've defined 10epochs now also during the training weevaluate the network predictions on thetest set which is basically the datawhich we haven't learned but this datais set aside for every fifth batch andthis is visualized so in our problemstatement what our network is going todo is it's going to predict the stockprice continuously over a time period oft+ 1 right we're feeding it data aboutthe stock price at time T it's going togive us an output of time t + 1 now letme run this code and let's see how closeour predicted values are to the actualvalues we're going to visualize thisentire thing and we've also exportedthis in order to combine it into a videoanimation right I'll show you what thevideo looks like so now now let's lookat our visualization we look at ouroutput so the orange basically shows ourmodels prediction so the model quicklylearns the shape and the location of thetime series in the test data and showingus an accurate prediction

(Editor in charge: code)

Previous article: Solidity, Blockchain, and Smart Contract Course – Beginner to Expert Python Tutorial - Ep39
Next article: GenAI Essentials – Full Course for Beginners - Ep101

• 
• 
• 
• 
• 
• 
• 
•