Time Series Prediction

Поделиться
HTML-код
  • Опубликовано: 3 ноя 2024

Комментарии • 169

  • @NickKartha
    @NickKartha 6 лет назад +11

    Thank goodness you did this! I needed this. I'm still doing that BTC bot from way back. It became part of my final year project seminar. RTA is super important. Thank you! Thank you for everything!

  • @radisadek7601
    @radisadek7601 5 лет назад

    Very clear (which is much much more difficult than people think)! Congrats! Two thumbs up!

  • @GZubatch
    @GZubatch 3 года назад

    one of the best videos i've saw !! Congratulations !!

  • @akrylic_
    @akrylic_ 6 лет назад +2

    Great video! VAR is insanely underrated and deserves more attention. Kudos

  • @jameslucas5590
    @jameslucas5590 6 лет назад +1

    I've been doing TS models since 2014. This was a nice summary and I am happy to see you plugged in Multivariant models.

    • @RV3ENS
      @RV3ENS 6 лет назад

      From your experience, what method would yields better results?

    • @ad98yt
      @ad98yt 6 лет назад

      What are your views about ARIMA?

  • @karameyer9499
    @karameyer9499 4 года назад +1

    Really nice video! Easy to follow even for a beginner! Thanks a lot from South Africa :)

  • @wolfisraging
    @wolfisraging 6 лет назад +11

    I've done tons of experiments on time series datasets and came to a conclusion, rnn models works best in any case specially GRU and LSTM.
    You can also have multilayered rnn structure with both GRU and LSTM layers(that was my idea, and it worked) embedded together. Also make sure your model is bidirectional

    • @ibozkurt79
      @ibozkurt79 5 лет назад

      Hi Rishik. Are you using interactions with other features like a supervised model or you just keep it as time series?
      I tried only with time series and did not get any better score with RNN. Thank you in advance

  • @mikashaw7926
    @mikashaw7926 3 года назад

    This is the best video, every other one was so confusing! Thank you!!

  • @agarwalamit081
    @agarwalamit081 6 лет назад

    Congrats on reaching half a million followers. The shirt videos are very inspiring to start learning ML algorithms

  • @dehanopperman6309
    @dehanopperman6309 6 лет назад +94

    I loved the overview and visualizations, but why skip the topic on stationary vs non-stationary time series and not mentioning AR, MA and ARIMA models before moving to exponential smoothing and LSTM networks?

    • @hfkssadfrew
      @hfkssadfrew 6 лет назад +11

      Dehan Opperman siraj is more entertainment guy rather than someone bring a course to you in 11 mins. And also, note that he didn’t mention stochasticity at all! so the time series he think is different from what you learn in the class. And he probably never took a time series class at all but it is fine. This is just entertainment, just to note that don’t take it serious.

    • @unclemax8797
      @unclemax8797 6 лет назад +3

      exponential smoothing car be described as a member of the "arima family"

    • @SirajRaval
      @SirajRaval  6 лет назад +24

      it was a question of time, i covered a lot in this short video. i'll make another

    • @dehanopperman6309
      @dehanopperman6309 6 лет назад +1

      @@SirajRavalThank you for replying. Just wanted to know if there was some specific reason. Time makes sense. Keep up the great work!

    • @raaghavsharma378
      @raaghavsharma378 5 лет назад +1

      Shaowu Pan which university you are doing data science class ?

  • @atineshs
    @atineshs 6 лет назад +1

    Was struggling to understand time series analysis, Thanks Siraj for an awesome explanation

  • @vaibhavvats2283
    @vaibhavvats2283 6 лет назад +5

    Just did time series prediction for Air pollution in Delhi for project, have to give presentation tomorrow.. This video is absolutely on time xD Explaining my model would be easier now!

    • @ArdeshirBanerjee
      @ArdeshirBanerjee 5 лет назад

      What did you use? I too have pollution data of a place.

  • @engineeringwithmehran
    @engineeringwithmehran 6 лет назад

    Love your simplifications of those complex concepts and use of media to visualize those concepts.

  • @ThePentanol
    @ThePentanol 3 года назад

    Thank you. It's short and to the point

  • @ukimalla
    @ukimalla 6 лет назад +3

    Love the going from simple to more complex approach! You should do that more on your videos. Unless you already do, and I was just too dumb to realize it. Keep up the good work!

  • @jacobtb1
    @jacobtb1 5 лет назад +6

    Dude. You are genius and I love you.

    • @darkquaesar2460
      @darkquaesar2460 4 года назад

      Oof

    • @ShivamPatel-ey9re
      @ShivamPatel-ey9re 3 года назад

      @@darkquaesar2460 man those complicated hilbert spaces are tough. Good thing we can do inside products on hilbert spaces

    • @darkquaesar2460
      @darkquaesar2460 3 года назад

      @@ShivamPatel-ey9re careful you might upset the jesus christ of AI

    • @ShivamPatel-ey9re
      @ShivamPatel-ey9re 3 года назад

      @@darkquaesar2460 dont worry. In my complicated analysis course we learned about logic doors so i think everything will be ok.

  • @kevinpunnoor
    @kevinpunnoor 6 лет назад

    Dude ... You are just nailing the industry... Damn good keep going.. was really thankful.. cheers..

  • @bambynaa
    @bambynaa 5 лет назад

    Siraj - This is a phenomenal explanation! Thank you so much. Your voice is clear and consistent. You are saving me... Love you!

  • @TheDistractionStudio
    @TheDistractionStudio 6 лет назад

    Thank You So Much. One of the best Times Series explanation I have watched. Appreciate it.

  • @Fawkesl
    @Fawkesl 5 лет назад +1

    Ty Siraj, you just help an CS Masters student in Brazil :D

  • @deneshkumarmani
    @deneshkumarmani 6 лет назад +1

    Improved on the sign language Siraj, that makes the video more comprehensible. Great work :)

  • @pcenxyz1838
    @pcenxyz1838 5 лет назад +1

    Thanks for boosting my confidence as well! The way you explain is extremely unique.Thanks a ton Siraj!😄
    Please can you do a tutorial video for weather prediction?And anyone reading this comment can you please suggest an application for the same?

  • @willembressers
    @willembressers 6 лет назад

    Hi Siraj. Have you ever tried prophet (from the facebook researchers)??. We use it daily in forecasting consumer sales succesfully. It facilitates multiple seasonalities (yearly, montly, weekly, daily, hourly) simultainiously combined with a trend and simple lineair regressor for the noise.

  • @rebelindianify
    @rebelindianify 5 лет назад +1

    Peter winters was Holts student who actually improved Holts exponential smoothing method to seasonality and it is called Holt-Winters method, not Holt’s Winter method. Sorry to be that guy. I thought this must be corrected. Thank you.

  • @arthurbarette768
    @arthurbarette768 4 года назад

    Is it possible to use a Ridge to make predictions of the future values of a time series, when we have problems of multicollinearity between the multiple explanatory variables ? Does it really prevents us from all the Time Series approach with stationnarity and all stuff ?

  • @MarceloNajar
    @MarceloNajar 5 лет назад

    Tremendous explanation! Just what i needed to make divergent thinking in TS

  • @aayushsingh3140
    @aayushsingh3140 6 лет назад

    thank you . really needed it for my project

  • @mudbath25
    @mudbath25 3 года назад

    This is pretty good!

  • @AZTECMAN
    @AZTECMAN 6 лет назад

    Great explanations Siraj. Very clear.

  • @solid84
    @solid84 5 лет назад

    Another great video! Thanks for putting this out there!!

  • @iyanu3723
    @iyanu3723 6 лет назад

    amazing post. please what clustering methods are the best for unsupervised learning of time series?

  • @MuhammadBilal-gf7ci
    @MuhammadBilal-gf7ci 4 года назад +1

    whole this lecture reminds me of my calculus, and algebra classes

  • @lukpisimoh
    @lukpisimoh 6 лет назад

    Awesome introduction! Can't wait to see the LSTM one!

  • @completelyboofyblitzed
    @completelyboofyblitzed 5 лет назад

    Amazing! Thank you!
    But could you make a video on how to define what kind of time series you have? Super short 🙏
    If you don't know that you have a moving average for example or autoregression, only your visualized series? The main features to define what kind it is

  • @vinayaktyagi8773
    @vinayaktyagi8773 4 года назад

    hey ,
    can we make multivariate time series data into univariate time series data by applying PCA or dimension reduction techniques ?????

  • @UsmanAhmed-sq9bl
    @UsmanAhmed-sq9bl 6 лет назад

    Awesome siraj. Loved your visualization. Need a follow up video. Please include facebook prophet ( a regression model for time series prediction) in your coming video.

  • @nomanshaikhali3355
    @nomanshaikhali3355 4 года назад

    Hey, I have a query in ARIMA MODEL of time series forecasting... I have given 2011 to 2020 data set to my model for stock values prediction but the model gives AIC values = 8813 and similarly, in the same algo, I have given only 2020 data from Jan to Oct for forecasting of nov stock values then the same algo gives me 789 AIC values!! Is Time series forecasting ARIMA algorithm works better on less data??
    Looking forward to hearing from you soon!!

  • @abramswee
    @abramswee 6 лет назад

    excellent video as usual. thanks for educating us.

  • @keenheat3335
    @keenheat3335 6 лет назад

    hmm the price data kind of look like a sinusoidal signal, wouldn't be better to predict the price using a sinusoidal function instead of a linear function? like approximate using a fourier series. IE: do a fourier transform on the price history to convert the time series data into frequency series data, then only take the term with major coefficient, then predict the data base a composite function that's made of those major terms.
    Although I could imagine you might lead to a lot of over fitting with this method.

  • @lakshaymalhotra1858
    @lakshaymalhotra1858 6 лет назад

    I have jst studied Time Series in this semester .. the actual application of theory is nt there in our college curriculum .. if Possible could you please share some links to cover time series from scratch to advance .. this video shows there are many things which should be tought to us .. bt can't rely on college for everything .. thank you Sir .

  • @nodarokroshiashvili3678
    @nodarokroshiashvili3678 6 лет назад

    Hi Siraj. What about some classical time series methods such as Auto Regression(AR), Auto Regressive Moving Average(ARMA), Auto Regressive Integrated Moving Average(ARIMA) and list can goes on

  • @mup3217
    @mup3217 6 лет назад

    which NN model do you think that can be good for predict the lottery result?

  • @NickKartha
    @NickKartha 6 лет назад +21

    5:35 "like a megazord"

  • @alvaromartin6301
    @alvaromartin6301 6 лет назад

    Thank you! This will help me with my thesis :D

  • @aiwebbiz8532
    @aiwebbiz8532 6 лет назад

    Great flow and visualizations... I am wondering how long it took(time) to create the video and research the idea, or what is the average time you spend producing a video?

  • @scott7948
    @scott7948 6 лет назад

    I am working on this problem right now. I used ARIMA for univariate time series analysis for real gdp forecasts. Keen to try LSTM but not too sure what other variables I could use since the influences on real GDP are forecasts as well.

  • @JordanHarrod
    @JordanHarrod 6 лет назад +2

    "Dark magic" definitely not wrong in some of the less interpretable models lol. Great video!

  •  4 года назад

    Random Forest Regressor is also doing a pretty good job as a baseline

  • @gtalpc59
    @gtalpc59 5 лет назад +1

    Hi Siraj, amazing effort! Thanka a lot. could you please video on feature engineering? Esp on time series. That would be helpful.

  • @abilashvr4802
    @abilashvr4802 5 лет назад

    @Siraj thanks for all videos you post. Could you please create an video on Multivariate time series with categorical data?

  • @NickKravitz
    @NickKravitz 6 лет назад +1

    Great video! I work in time series forecasting; we use a lot of ARIMAX type models, similar to your VAR model, allowing AR, MA and covariate regression terms into the model. We still can't predict the stock market; it is model resistant :(

    • @mikenotangel
      @mikenotangel 6 лет назад +1

      Nick Kravitz he missed also variance modeling with GARCH models, and other multivariate models such as Vector Error Correction

  • @tuxmania
    @tuxmania 4 года назад

    Question here: So Holt-Winter is able to predict the future multiple steps ahead, so to say if I fit Holt-Winter on 2010-2018 data, I can probably predict all of 2019 pretty well (all monthly). But with LSTM models I have the issue that it only predicts the very next data point (in my example Jan of 2019). When I set multiple neurons as output layer in my (stacked or not stacked) LSTM network I get very bad results. So is there a possibility with neural networks to predict multiple steps ahead like ARIMA and Holt-Winter can do it?

  • @indiansoftwareengineer4899
    @indiansoftwareengineer4899 6 лет назад +1

    Love you sir and also this channel.

  • @giorgostsilivis6771
    @giorgostsilivis6771 3 года назад

    10:28 we all came for this!

  • @shalinianunay2713
    @shalinianunay2713 5 лет назад

    Good one.

  • @jinwu3262
    @jinwu3262 5 лет назад

    This feels like what I have learnt in my supply chain course.

  • @GuillaumeChevalier
    @GuillaumeChevalier 6 лет назад

    Hey this is the figure I created at 9:54 into the video! :)

  • @hrvaticaigitara
    @hrvaticaigitara 5 лет назад

    good video, thanks!

  • @abhishekprasad5359
    @abhishekprasad5359 5 лет назад +5

    Siraj: I guess this video should be split further to cover ARIMA, ARCH,GARCH,EGARCH, VAR etc. Unfair to cram so much info by skipping others as timeseries is vast topic.

  • @akshaynautiyal6644
    @akshaynautiyal6644 6 лет назад

    Amazing. Thanks for summarising this !

  • @patilpv123
    @patilpv123 6 лет назад

    Siraj will you pls guide me on how A MECHANICAL engineer will learn ML and AI

  • @tyler_russell
    @tyler_russell 5 лет назад

    Anything new in the time series prediction space over the last year?

  • @DerKinGGonZo
    @DerKinGGonZo 6 лет назад +3

    I wrote my master thesis on multivariate time series prediction with a focus on parametric vs non-parametric methods. What you just easily explained in 10 mins fill nearly 100 pages in my thesis. Still you skipped A LOT what I discussed. BUt still nice overview.

    • @klio1095
      @klio1095 5 лет назад

      Hi there! Would you mind sharing your thesis with me? Was thinking of choosing the same topic for my bachelor's thesis too. You could DM me seperately if you want😀

    • @FabioTMW
      @FabioTMW 5 лет назад

      I would also love to read it, i was thinking of doing something similar in my master as well

    • @ibozkurt79
      @ibozkurt79 5 лет назад

      +1

  • @oscarbecerril8343
    @oscarbecerril8343 6 лет назад

    Your videos are awesome as always, Siraj. But please don’t move the text so much, neither the equations

  • @dimi3815
    @dimi3815 6 лет назад

    Which editing software do you use?

  • @krisschobert4484
    @krisschobert4484 5 лет назад +3

    You're delivery is great and video editing is excellent, but some of the plots and equations are misleading and don't align with why you're saying. For instance, you mentioned how the winter clothing store has more sales during winter, but your plot at 7:37 shows the opposite. Please be really careful with your equations and plots. Dialog helps us along, but the plots/equations are what solidify our understanding.

    • @TheDodito
      @TheDodito 5 лет назад +1

      Excellent observation.

  • @abhishekprasad5359
    @abhishekprasad5359 5 лет назад +4

    Personal I would like to see a video where you talk about predictive modelling using PANEL DATA.

  • @alvaromartin6301
    @alvaromartin6301 6 лет назад

    Can you say some time series prediction for long term like 20 years?

  • @unclemax8797
    @unclemax8797 6 лет назад

    garch models, and their extensions ( including their multivariate developments) have more practical applications in finance than exponential smoothing methods ( that are ok for easy to solve problems)

  • @mikenotangel
    @mikenotangel 6 лет назад +5

    You didn’t mentioned the kings of time series modeling such as ARIMA, ARFIMAX, SARIMA models in order to model mean, and to explain variance GARCH models in general

    • @hrvaticaigitara
      @hrvaticaigitara 5 лет назад

      As a beginner, I appreciate that you pointed this out! I know what to google :D thanks

  • @_akshaydeep
    @_akshaydeep 6 лет назад

    i have few gbs of time series data. i want to store it in db. but MySQL is kinda dumb for that. pg and timescaledb looks promising to me. any suggestions on which db should i use?

  • @rakeshmallick8040
    @rakeshmallick8040 6 лет назад

    Very useful.

  • @excel_wang
    @excel_wang 5 лет назад

    Not sure about the last part since NN such as LSTM cannot even predict better than a random walk model.

  • @Tuguldur
    @Tuguldur 6 лет назад +1

    4:03 The best head shaking

  • @nikhilwadibhasme9608
    @nikhilwadibhasme9608 6 лет назад

    Seasonality in stocks covered news correct??

  • @agnibga
    @agnibga 5 лет назад

    I didnt understand the error graphs....like how we checking errors...can you explain

  • @dhusor9633
    @dhusor9633 4 года назад

    how can i forecast multiples item forecasting ?

  • @WillTesler
    @WillTesler 6 лет назад

    I've had success doing time-series prediction which just a layered dense NN.

  • @udayaravedu9068
    @udayaravedu9068 6 лет назад

    can you please make a video on deployment of mask rcnn

  • @suryabhusal1527
    @suryabhusal1527 5 лет назад

    What would be accuracy with fourier transform?

  • @scottmiller2591
    @scottmiller2591 6 лет назад

    That should be Holt-Winters, not Holt's Winter. Holt and Winters were two different people that worked on this filtering technique; the fact that it is seasonal and the 2nd author's name is Winters is just a happy coincidence. See Rob Hyndman's chapter on Holt-Winters here: otexts.org/fpp2/holt-winters.html#ref-Winters60

  • @modelworkzseo
    @modelworkzseo 5 лет назад

    Why didn't you use your timeseries to predict your copy / paste scandals?

  • @patilpv123
    @patilpv123 6 лет назад

    Also Share the opportuintes in manufacturing sector for ML

  • @MrZouzan
    @MrZouzan 6 лет назад

    What about Support vector Regression ?

  • @SomeRandomIdeas
    @SomeRandomIdeas 5 лет назад

    the ever going gamble...is this also stolen or not?

  • @RandomGuy-hi2jm
    @RandomGuy-hi2jm 6 лет назад +2

    i have tried this before publish of this video but failed..

  • @adip8
    @adip8 5 лет назад

    Is this video for engineers since they hardly know any math?

  • @alexbadea5594
    @alexbadea5594 6 лет назад

    Very nice!

  • @thegoodbetdotcom3069
    @thegoodbetdotcom3069 5 лет назад

    I make a profit almost every day trading and if I can get to the point of using a neural net of some sort, probably along the lines of RL with some time series concepts then I will be making millions from the increased signals. It's only a matter of time to find the right way to get help on this without the secrets ending up on github, lol. Most CS people (I'm not one) I read about trying to apply ML to financial trading just don't know what to do as traders, it's not the algo's that is the problem it's not understanding the environment. Anyway, one step closer from watching this I guess.

  • @stefan-ls7yd
    @stefan-ls7yd 6 лет назад +4

    Thought the video was about predicting T-Series subs vs Pewdiepie ...

  • @pink814floyd
    @pink814floyd 6 лет назад

    you can't predict non-linear models; they don't teach this at ivy league schools?

  • @aiacademybysid5631
    @aiacademybysid5631 6 лет назад

    BTW love ur video Big fannn

  • @lennar10
    @lennar10 5 лет назад

    Anyone found a Python library that would get VAR models done efficiently?

  • @tamalsaha8837
    @tamalsaha8837 4 года назад

    I'm looking for an exact equation for "Linear Regression Extrapolation on Time-series spatial data". can anyone help me?

  • @ItzGanked
    @ItzGanked 6 лет назад

    good educational material, real world application probably not

  • @mariuskuyler7763
    @mariuskuyler7763 5 лет назад

    This is fucking gold

  • @ПопулярновБългария

    how can i download data from different indicators ?

    • @itswasLyric
      @itswasLyric 6 лет назад

      I'd check out kaggle data sets, and I think some sites has rss feeds or an API for all of their articles... if youre working with something that news could be a variable that is...

  • @markfelstead4586
    @markfelstead4586 5 лет назад +5

    Take away the "ime" from Time Series.
    What do you get?

    • @xordux7
      @xordux7 5 лет назад

      We get the most subscribed youtube channel :D

  • @HarvinderSandhuEsq
    @HarvinderSandhuEsq 5 лет назад

    It is “Holt-winters” method

  • @venkateshr6031
    @venkateshr6031 6 лет назад

    Your regular fan here Siraj! Would love to have a chat on your Slack channel.