Lets Implement LSTM RNN Models For Univariate Time Series Forecasting- Deep Learning

[@souravdey1227]

One of the finest explanations in youtube. Krish Naik is a legend

[@shekharkumar1902]

Amazingly simplest way complex concept of LSTM explained ...you are getting many blessings from aspiring Data scientists . Thanks a lot

[@kvafsu225]

I must say it is an amazing presentation on LSTM. Thanks a lot.

[@hassanayaz6242]

Professor Amazing?finally learn through your video.useful time to see your video

[@BhupinderSingh-rg9be]

Was waiting for this from long time. Please sir more videos on deep learning. Great explanation by the ways

[@prakashdhakal8308]

Thank you explaining this in simple manner . Waiting for multivariate time series analysis

[@Hitesh-Salgotra]

Krish sir i ever get a chance to thank you I will surely not gonna miss that chance you always save my ass from errors bugs and help me to understand in a better way

[@advaitshirvaikar6101]

Great tutorial! The n_features you talked about is the number of features in one time step. i.e. if x=[[1,2,3],[4,5,6],[7,8,9],[10,11,12]], samples=4, time_steps=4, and n_features=3

[@WahranRai]

You must re-read Jason Brownlee' sblog in order not mixing terminoly ! Your example is univariate time serie nb_features = 1 and what you call nb-features (preparation of data) is the nb_step !

[@jordanle6004]

Thanks!

[@sidheswarpatra3732]

Thank you for explaining it in such simple manner. Eagerly waiting for the multivariate Time Series analysis and it will be great if you can work on some live data and demo us with these concepts such as Stock price forecast, Cryptocurrency prediction or any other real-life usecase.

[@helpyburhanuddin3087]

I salute you sir for your explanation and choosing the concept.

[@Chandrashekhar2006-z3f]

Hello Krish sir, Your Machine learning playlist was great enough to clarify many doubts for someone like me who is aspiring to be data scientist. Can you make playlist on time series analysis please! :)

[@b0nnibell_]

please make a video for multivariate time series forecasting....your explanation sir is awesome!

[@touriafransform6280]

Hi Krish, 5:11 You are converting the time series into 'supervised' data (X, Y). I have a few questions and I hope you can answer them. I always thought recurrent networks could be trained without 'Y' because they can detect "temporal patterns" in X itself. 1) What advantage do LSTMs have over normal deep neural networks or other regression algorithms because you can fit a regression between Y and X. 2) Why is it necessary to convert into 'supervised' data (X, Y) before training LSTM? Time series models such as ARIMA do not require converting into supervised data. They can train by 'swallowing' the whole time series. Can't LSTMs do this? 3) Any machine learning models that can predict time series like ARIMA without converting into supervised data?

[@HELLOCHAND100]

Any answers to these queries?

[@rakesh2you]

Krish a very nice and good explanations for the topic.Thanks for spending your time and helping.

[@ramdhanandtara]

Thank you for the explanation, really help me for complete my thesis

[@kabeerjaffri4015]

You are the type of person i wanna be. Great video krish love from Pakistan

[@kashifjavedlone1780]

Great tutorial! Please also make a tutorial on Multivariate Time series forecasting

[@uditarpit]

n_feature will change based on number of other variates. Basically third dimension in input data is 1 here. (n_feature). creation of lags will be same.

[@premaldoshi9824]

Hi sir, eagerly waiting for the multivariate analysis video

[@adityataksande5024]

Thank you very much sir or this i was waiting for this so long.

[@sanjeevkumar2004]

Thank you so much Sir, please also do a multivariate time series where there are few categorical and numerical features which do not depend on time. I have one problem which I could not solve wherein the data was given for 24 days on hourly basis and prediction was supposed to be made for balance 6 - 7 days (hourly prediction) for each month. it is like information of last week of each month for a span of 4 - 5 year is not present in train but present in test for which prediction has to be made. I would appreciate if my comment catches your kind attention. Thanks for your continuous support and helping many aspiring Data Scientists out there.

[@ragulshan6490]

A little piece of advice, just try to learn from scratch like basics of time series components such as ETS and ETS Based models then try out more difficult methods of time series like arima, SARIMax, varma etc.. Please don't try to apply lstm model blindly whenever u deal with time series problems. You can do wonders with simple and traditional models!

[@sanjeevkumar2004]

@@ragulshan6490 yes I understand. I just wanted to find a solution to it. Not that am using LSTM for the same. I am aware of those models to some extent ☺️ thank you

[@ragulshan6490]

@@sanjeevkumar2004 Try udemy course from jose portilla's Time series. You'll learn a lot

[@ranjanpal7217]

Amazing explanation...Plz make a video on intuitive idea of how LSTM works for time series prediction

[@syedaamir3238]

excellent teaching sir...so kind of u!

[@rajeshahir_gj6765]

Thank you for the detailed implementation of LSTM - RNN. My question is how do I utilized the LSTM-RNN for unsupervised time-series classification?

[@BhaargavL]

Thank You Krish for the excellent demo!

[@sruthiparvatha8790]

Thank you so much for this, explanation was great and understood it very well. Keep up the good work.

[@avinashchandane7464]

Thanks a lot.....👍 .very helpful for me...!!

[@augustedalby8058]

Thank you for your explanation, it is so simple. and easy to understand! i hope you will give us more videos like that, it is so helpful!

[@nazmulshohan8807]

Sir, Take love ...Great explanation...

[@W.M.J.M.Welagedara-Universityo]

You're superb man 🤩🤩

[@suvarnadeore8810]

Thank you sir for amazing explanation..... and happy Teachers day sir..5/9/2020

[@JKim-p1r]

Very clear and thank you for explaining this

[@dzifahodey7214]

Very clear explanations. Thank you!

[@AI_Adhyayana]

Hi Krish, nice video . My question is what if there is seasonality in timeseries will LSTM be able to capture it.

[@malindal4u]

don't you have a video for Multivariate Time Series Forecasting With LSTM Complete Python Tutorial

[@sandipansarkar9211]

Finished practicing in Jupyter notebook. Thanks

[@manojsriramula2355]

looking for the multi variant forecasting !!

[@sidharthadaggubati438]

Hi krish , wonderful explanation as always. Just a small doubt, In prediction logic, instead of appending day wise predictions and again feeding them to model.predict , is it possible to do batch prediction for 10 days at once?

[@siddheshmunagekar9915]

Thank you so much Sir, Nicely explained and is very helpful

[@k33jakha]

Thank you Krish! it helps prety much to understand however can you be able to make a video using GRU (Gated Recurrent Unit) on the same dataset!!

[@raghuraman.kkoteeswaran.v229]

very well explained sir.

[@anurag040891]

Great Video sir. Please make one video on chat bot chat analysis .. or if you have please share the link.. It will be very helpful for me....

[@rahultrivedi4138]

What shoulde be ideal time step for data volume. Can you pls explain? Because with the change in time step the output changes

[@rubikambo4927]

if we have missing data in timeseries data so what are the various way to handle that or null data.

[@sandipansarkar9211]

Great explanation. Need to get my hands dirty with practicing in Jupiter notebook. Thanks

[@abhisheksrivastav1737]

Hello Sir In below code while forecasting next 10 values , code is not working and giving error TypeError Traceback (most recent call last) in () 1 import array 2 # demonstrate prediction for next 10 days ----> 3 x_input = array([187, 196, 210]) 4 temp_input=list(x_input) 5 lst_output=[] import array # demonstrate prediction for next 10 days x_input = array([187, 196, 210]) temp_input=list(x_input) lst_output=[] i=0 while(i3): x_input=array(temp_input[1:]) print("{} day input {}".format(i,x_input)) #print(x_input) x_input = x_input.reshape((1, n_steps, n_features)) #print(x_input) yhat = model.predict(x_input, verbose=0) print("{} day output {}".format(i,yhat)) temp_input.append(yhat[0][0]) temp_input=temp_input[1:] #print(temp_input) lst_output.append(yhat[0][0]) i=i+1 else: x_input = x_input.reshape((1, n_steps, n_features)) yhat = model.predict(x_input, verbose=0) print(yhat[0]) temp_input.append(yhat[0][0]) lst_output.append(yhat[0][0]) i=i+1 print(lst_output)

[@augustedalby8058]

Again, thanks for your explanation, but I have a question. would it predict the future for a range of dates?

[@minudixit9956]

Hi Krish, I have been following your videos since one of your first 10 videos came out. I have seen your journey from Data Science learner to expert, it is so motivating that too in such a short period of time. To implement a new topic into a problem statement I first watch your videos, it gives a kick start to the journey. My query regarding LSTM for time series, what thumb rule we should follow for setting the Lookback/window size ?

[@rajasaroj2297]

Thank you soo much for this awesome explanation :)

[@J0e65]

Thanks for the tutorial. But after calling the model again and again, the output results are degrading. How to apply random seed to the datapoints, so that I get the same test results every time on the same dataset?

[@rubikambo4927]

what attributes we can use in time series analysis or for prediction can u suggest me.

[@rubikambo4927]

thanku sir i m using this model in my work can u plz tell me is CNN beter or LSTM for time series data related to price . any other dependent data on that particular time can we use it here paralley to the model. n u used the stacked lstm model na. Initialy inmy phd work i m not able to work with model but now able as i had seen jason mastery blogs i also strted my work from viewing blog of jason n u explained well sir. thanks ur video is helpfulful for me.

[@tintumarygeorge9309]

n_features = 1, this value denotes the number of features in the input, in this example, it is univariate forecasting so only one variable/feature . that is why n_features = 1

[@shreyaskulkarni5823]

You could use recurrent activation='sigmoid' as well ...the loss value is extremely less with recurrent activation

[@anokhikumari6900]

plz also make a tutorial on how to tune the training optimization algorithm

[@zaynahchummun668]

how do we know how many days should we use as independent variables?

[@محمدمحمود-ت7ح4ظ]

many thanks for this tutorial very helpful. but i am little confused about the difference between this video and the Stacked LSTM- Deep Learning video it feels very similar

[@muhammedcansoy1434]

Excellent explanation, thank you

[@rohitdas6515]

Hi Sir. I want to evaluate my model (not through graphs) . also, I have done mean absolute error and rmse. I cannot take mape because my actual target as zeroes in it. Kindly mention more ways to evaluate model

[@nagashishsv843]

sir is this the continuation of complete machine learning playlist or i should see deep learning playlist before seeing these videos

[@krishnaik06]

Please see the deep learning playlist first

[@swapniljena8684]

Is the 3rd dimension added to indicate the dimension of data? Like as it was only 1D then we used n_features = 1. If we used it for 2d matrices prediction like in images we can use n_features = 2. Pls correct me if I'm wrong, I'm unsure.

[@Vish_27-v8x]

Please show us how to do multivariate LSTM and also tell us which parameters we must tune Incase we have a very high RMSE. Thanks in advance.

[@sreeramsaravanan8132]

@krish please put a video on Simple exponential smoothing

[@vedantsavant4933]

'module' object is not callable error in colab while predicting 10 day data

[@عدنانمهداوي-ن5ث]

why we use 3 previouse value not 4 or more or less ??

[@MrsDyarvane]

AttributeError: 'array.array' object has no attribute 'reshape'. I am getting this error before i could predict the 10 values

[@malice112]

Can you please do an example of an LSTM model using a Functional API architecture and not Sequential()

[@arunthandra5065]

Hi very good explanation... when I run this code I'm getting the initial 'yhat' of 224.32 instead of 222.60, and because of that all other output varies. Can someone tell the reason please?

[@hardikshah4304]

Is this video included in Machine Learning Playlist or Deep Learning?? Your Videos are amazing Sir. Hats off !

[@yathishl9973]

Hi Krish, It was a wonderful explanation you are an asset to Data Science. I have a query here, which was asked in one of the interviews as well, please suggest. If LSTM is built using 'sigmoid' and 'tanh' activation functions as explained in LSTM intuition why are we using 'relu' again which building LSTM as shown in the video, shouldnt we using either sigmoid or tanh?

[@lakshminarayanans1460]

Bcuse we want output as continuous variable where as sigmoid or tanh will be used for categorical variable.

[@aldakurniaputri_fmipa6050]

Why didn't you divided the data into train data and validation data?

[@md.jeweldhali1965]

thanks a lottttt.... its really helps me

[@naveennoel9496]

I am very thankful to you, but it would be very helpful if you could take random distributed data and given insights on that

[@geniuslearningsarat3392]

Hi Krish, Please help to post any vdo on forecasting sale for multiple products in one go using any time series, LSTM,FBPROPHET or SARIMAX etc.

[@TheMartin68261]

Any Books recommended for LSTM time series forecasting

[@Nandeesh_N]

is this the last video in this complete ML playlist?

[@tedtalks2144]

Can u do for multi task also

[@Nao-Tomori]

Why must we reshape x into 3 dimension?

[@harisramzan9814]

Please upload multivariate timeseries with LSTM and ARIMA..

[@heshamgaber94]

Why don't you normalize the data when fitting into your lstm model? And also why did you use relu activation function, as the literature used sigmoid!

[@solano.todeschini]

Sigmoid activion are used for classification problems, whereas relu activation will give you a continuous value as output, making it suitable for regression problems - the same kind explored in this video. The same goes for normalization - if you'd like to make a classification, that'd be better to normalize and scale all data so you can expect a better loss. For regression models, normalizing and scaling would give you the wrong outputs since it pushes data out of its natural range...

[@riagoel2885]

#krish sir please make video on multivariate analysis

[@uditarpit]

n_feature is important but why is missing ? where will n_feature not be 1 ?

[@adegboyegaadebayo1080]

why is yhat variable indexed in both one and two dimenssions

[@SaurabhSingh-ex6vl]

Hi @krishnaik06 i want to know how to create 3d array which has multivariate and (multiple targets example like 2).

[@niazmorshedulhaque4519]

Dear Sir, Thank you for your nice tutorial. Could you please make a tutorial where we use both prediction models : ARIMA and LSTM for same dataset.

[@ratrat2056]

hello, I trying make you code but allways have error on - x_input = array([187, 196, 210]) IndentationError: unexpected indent

[@mariochavezpazmino3152]

Me too, What did you do ?

[@DeepFrydTurd]

;0 are you running this on a TPU GPU OR CPU?

[@swanandkulkarni126]

Is this the only way to construct LSTM TSF model?

[@pankajkumarpandey8615]

sir i want to join your project live group. how much project solve in 1 month?

[@VenkateshDataScientist]

why did you go back to 3 values

[@AshwinMaccount]

Hi sir, Can you do the same lstm from scratch with no frameworks , from scratch. Please each step by step by while you implement the lstm from scratch. I 've searched everywhere but no video is found. Please sir

[@gauravlotey4263]

Sir please me ek international student project pe kaam kar raha hu. I am in dire need of help. Can I please connect you if possible. Sir mere pas help source nahi h koi, I strictly follow your teaching. Please help me in this time series forecasting challenge

[@cablemaster8874]

Hello, Dear Krish Naik, I have put my data in available codes, but one problem is coming, please solve it NameError Traceback (most recent call last) in () 1 # demonstrate prediction for next 10 days ----> 3 x_input = array([71,70,93]) 4 temp_input=list(x_input) 5 lst_output=[] NameError: name 'array' is not defined

[@muhammadasadaslam3991]

try changing it to Numpy Array using np.array

[@samiyamohammedsulimanalshu8516]

please make video for cyber attacks classifications using lstm

[@justinhuang8034]

Damn man great stuff like always

[@kunal7503]

Amazing video

[@RajRajkolhe]

Why value are not taken randomly

[@shoaibakhtar5889]

You were going great but at the last part of prediction you did wrong, you have to explain the prediction part as well...Otherwise it was a great tutorial

[@mdsaif831]

Sir when yur new project is coming for member

[@krishnaik06]

In a couple of days