Time Series Forecasting with Facebook Prophet and Python in 20 Minutes

[@helloonica8515]

This is by far the best tutorial video, you went straight to the point and you were able to explain everything properly.

[@joao_ssouza]

I'm about to start a project at the university related to time series forecasting, and you helped me a lot, thank you very much.

[@macewindont9922]

As a newbie to forecasting, it helped a lot that you went slowly through all the pandas and prophet api calls.

[@NicholasRenotte]

Glad you enjoyed it @Marcel!

[@lukasmendes4625]

I take my IBM courses, but after I always come to your channel to see your videos as they give me a much easier understanding. Thanks for this, and great content as always!

[@mohammedzain9876]

Just an update to people watching this video in 2022 if you get an "ModuleNotFoundError: No module named 'fbprophet' " its because the package name changed to prophet, so if you do - from prophet import Prophet - that should work!

[@keivanmokhtarpour4863]

One of the best videos I've ever seen on KZbin, with maximum information in minimum time!

[@keivanmokhtarpour4863]

I only went through the code without listening to your voice :D

[@titaniumwolf2]

Cheers bro. I'm a web dev but suddenly have to so something like this. Awesome teaching skills.

[@iliovininino]

Great job!! So far the best I've found explaining prophet. There is no full course yet anywhere... I mean, explaining prophet's hyperparameters tunning, and exploring the tool in more detail.

[@shyamjaiswal7114]

You got a new subscriber from India.

[@berkceyhan5031]

Great video for beginners! Thank you for explaining every single thing without being boring. I enjoyed and learnt at the same time. Thanks.

[@sarahkadi8115]

Thank you so much, i've never watched a video with someone explaining this way, you dind't forgot about any detail and it's perfect for people who begin! thank you so much !!

[@shanenicholson94]

Nicholas, this is the best tutorial I've seen on youtube...great work buddy.

[@LeandroMartinzzz]

wow!!!! Thank you so much. You speak very clear and explain all the steps. Great video

[@martinthabang9621]

This has been so helpful. I was already reaching my frustration limit. Thank you sooo much

[@jawadhassan4917]

I am impressed by the way you plan and execute well done.

[@Eysh2009]

This video is BEAUTIFUL, it helps so much! Thank you for the top quality tutorial!

[@jaeen7665]

This is how a tutorial should be done. Liked, commented, and sub'd.

[@Foundnoidentity]

I would really love to thank you so much, you explained it so well and I am finally able to forecast using prophet after watching so many other videos!

[@Dogcat677]

Best KZbin explanation by far so clear, easy for beginners to follow 💯💯

[@samm9840]

Thank you very much. Can you share how we can do validation for such time-series models once developed?

[@pavankumaravn5493]

Great video. explained the forecast model in a simple steps.

[@Zzzkkk1313]

Hey Nicholas. thanks for the video. could you please show how to do it with multiple products?

[@NicholasRenotte]

Yup, think I'm going to do a full tutorial on end to end sales forecasting!

[@merimagdi]

So much value here! Thanks! You got a new subscriber. Hi from Spain!

[@NicholasRenotte]

Thanks so much @María, much love back at you from Spain!

[@ermiasdejene]

Thanks, bruh. It was simple and straight to the point tutorial. Loved it. And your presentation was clear as well as your summary with identifying the overall flow of logic was epic. God bless you, bro.

[@guannanliu9999]

Your datetime doesn’t have time of the day, how did you get daily seasonality then?

[@JoseGutierrez-in6bn]

Your totorial is amazing, Congratulations you are the best.

[@mohitpande2006]

Hi nicholas, I am getting prediction output as date (1960-01-01T00:00:00) but I only want date not time is their any way out.

[@NicholasRenotte]

Can change the date format using this function: www.programiz.com/python-programming/datetime/strftime

[@transform2532]

@9:03 can't we just convert the datetime column using pd.to_datetime(df['Time Date']).. instead of four lines of code?

[@aminaleali7161]

Good Video. There was no time column. How did the breakout show the distribution with time as its x axis?

[@Benny65436]

Thank you again for the helpful video. What I don't understand are the numbers in the trends. For example, at 17:54. What does the -30 on Friday mean? We can't sell minus 30 products. Is it the deviation from the "standard"?

[@vamsikrishnabhadragiri402]

Thanks a lot for your video, what if we have different product names(let say 4), and stores(let say 2) and predict the value. can we still use Facebook prophet or do we need to build different models, which means 4*2= 8 models separately?

[@NicholasRenotte]

Build multiple models, I show it here (I screwed up a bit during the stream but the theory is the same): kzbin.info/www/bejne/rYm2anywec-okLM

[@MaxGroßeHerzbruch]

is it possible to look at the final model in an algebraic form? Like forecast= 4,3*weekday + 2,1*weekday*seasonality -1,234*seasonality?

[@zaynaba6626]

are you able to use Prophets to forcast bitcoin price using twitter sentiment? Would love to see a video on that!

[@tanmaykatke2611]

What to do, if I have multiple features? Should I plot them together? Or individually?

[@ankushpandita7548]

great man!! You explained it so clearly. Very Helpful

[@NicholasRenotte]

Thanks so much @Ankush!

[@AIandVisuals]

Very good presentation, but where is the train/test split, the cross validation, and the model evaluation?

[@nickt423]

@13:58 all of the rows in my yhat column seem to be about +4.13 different than yours

[@NicholasRenotte]

Heya @Nick, are you running the forecast on the same data?

[@nickt423]

@@NicholasRenotte I believe so, yes

[@NicholasRenotte]

@@nickt423 the output will likely be slightly different when you train it. But if you're getting wildly inaccurate results when there mgiht be something we need to dig into.

[@nickt423]

@@NicholasRenotte Okay gotcha. Yeah the output is very similar to yours. Nothing really to worry too much about. Anyways, thanks for the video. Subscribed!

[@NicholasRenotte]

@@nickt423 ah got it! Thanks so much!! Anything else you'd like to see?

[@SyedShakilAhmed-o7i]

What to do if there are more SKUs and different shop locations?

[@prasadseptember]

Hi, Thank you for sharing this wonderful lecture How can we build a model that handles millions of time-series data, like customer forecasting Please share your thoughts

[@NicholasRenotte]

Check out the data science dojo channel, I did a collab with them where I did something like that!

[@gistech8777]

Tnx @Nicholas! is it appropriate to implement this forecasting method in a data set that has date/time value but not a daily reading. for example incident data like traffic accident?

[@miguelpereira9095]

Great video, is it possible to update the model in a sliding window way?

[@rayantalwar8315]

Nice video! I have a question. In your video why does prophet forecast current values as well? Like the values for 2018 are already present and when we run forecast.head() why does it display different values for those 2018 dates?

[@sehgalkarun]

As you said We can make a product-specific time series But let's say I have 1500 stores and each store is selling 2000 products then how to tackle this ?

[@NicholasRenotte]

Loop through each combo. I'm doing a webinar with Data Science Dojo on this in a few weeks time!

[@sehgalkarun]

@@NicholasRenotte also you have removed store al well as product and only keep date and value ... But in real life I need to know the forecast store and product wise.

[@NicholasRenotte]

@@sehgalkarun no problemo, I'm doing a webinar with @DataScienceDojo soon on how to scale it up!

[@spatialnasir]

Thanks. A lot clearer than the official docs.

[@Tredetion]

This is very useful towards my masters! Thank you so much!

[@abstractnonsense8344]

Maybe I missed it, but did he do a hold out?

[@danielholocsi440]

Hi, how do I forecast for different product within different stores?

[@zacharygreenberg1831]

When you run timeseries with FB Prophet, do you have to stationarize your data, or will Prophet do it for you?

[@NicholasRenotte]

Heya @Zac, I don't normally perform any preprocessing (including stationarizatio) on the data before passing to Prophet and normally receive reasonably performant results. I'd run without it first and see how you go!

[@dominicatuahene7303]

amazing tutorial Nicholas. thank you so much. do you have a tutorial on a multivariate prophet forecast

[@juanmoctezuma9225]

Awesome video Nicholas! your explanation did help me to build a model that I need for my personal project, muchas gracias!

[@NicholasRenotte]

De nada, thanks so for checking out the video @Juan!

[@edwardsamokhvalov6720]

Daily seasonality is for intraday seasonalities, but you do not have intraday data so why would you specify it to true? It won't be able to generate intraday seasonality from eod data. Or am I not getting something???

[@AJ-ks8iq]

thanks! I like the style. can you do one for airlines sales where 2020 had a negative dip. and also focus more on the data science aspect of the data.

[@NicholasRenotte]

Heya @Anita, sure, I'll add it to the list!

[@AJ-ks8iq]

@@NicholasRenotte thank you Nick :)

[@NicholasRenotte]

@@AJ-ks8iq you're welcome!!

[@mooncake4511]

pip install fbprophet is erroring out in VScode windows. Any work around ?

[@NicholasRenotte]

Got an error for me?

[@senarkit]

Would be great if your video volumes are higher. (I am at my MAX and still have a challenge listening to you w/o headphone) But great video, thanks a lot Nicholas. Please keep making more videos on forecasting that also covers HYPERPARAMs and tuning them.

[@leoeveee]

What if we have missing dates in data, like no data for weekends

[@spider279]

hello Nicholas , how to do hourly forecast ( my ds is by 15minutes interval and my y is temperature and i want to do 3h forecasting of temperature ) please help me

[@nitishmc6929]

Hi.... I'm getting error" no module named fbprophet....how to resolve... please help me

[@NicholasRenotte]

Heya @Nitish, might need to install it !pip install fbprophet

[@charlesnwevo2706]

Great content, thanks a lot it was very easy to follow your explanations. Quick question, I was wondering if prophet has any metric for calculating error assuming I want to compare it with a different model?

[@kepenge]

Hi @Nicholas, Are you using M1 or Intel based Macbook, and what version of Python did you used in this tutorial?

[@tinashemuzata2159]

Hi Nicholas . Thank you for the video. Just a soft issue why do the *yhat* values differ from some of the historical data points.

[@tlghnkck]

I'm getting a "Time Date" error from the csv file I downloaded from the investing site. How can I fix this?

[@yenchu28]

Thanks, this gives a good start. Would be good to show how to add confounders and show interactions between different products if there are indeed associations, rather than having multiple univariate predictions. Also can show how to regularize and dealing with underfitting as it seems to do with a simple model.

[@SannidhiPHebbar]

great video Could you please explain forecasting when there are multiple features and multiple product store values

[@fahadabdullah510]

Can you please make a separate video on which is the best model for time series like LSTM,Darts,ARIMA,SARIMAX,FbProphet by giving some examples. Thank You

[@yousefpanahi7911]

hi, can you give me the link for the data you used in the course?

[@NicholasRenotte]

Dataset's in the GitHub link in the description :)

[@theflyingpharmacist4094]

Thanks for the great video. Do you know if you can add parameters 1) to set a daily max i.e if you know now more than X units can be sold per day and 2) set total number of units for sale i.e. limited edition merch with only 25m to sell? So it would stop at that point?

[@NicholasRenotte]

Heya @TheFlyingPharmacist, you could apply your maximum limits to the yhat column using something like this, change the value in maximum_units to apply your hard stop: maximum_units = 25 forecast['yhat'] = forecast['yhat'].apply(lambda x: maximum_units if x>maximum_units else x)

[@sagarwadile446]

How to Deploy of Gold_data. this fbprophet model in Pycharm using streamlit. Please Provides codes or Video

[@AndrewMoMoney]

Hey! nice production and editing, the code is nifty as well

[@NicholasRenotte]

ANDREWWW! 🙏 thanks so much man!!

[@cesareme]

Hi! Good Job! I've a question, maybe you can help me. My dataset contains 24 clients and 20 products, how could I run this code to calculate the forecast for each combination client-product-month? Thanks in advance!

[@NicholasRenotte]

Check this out: kzbin.info/www/bejne/rYm2anywec-okLM

[@cesareme]

@@NicholasRenotte Thx Bro!

[@Daxter296]

Thanks mate, I'm glad you explained each part really well!

[@Dogcat677]

Just curious is there a way to continuously input daily data and continuously predict future data ?

[@Tredetion]

Can Prophet take into account multiple variables that might affect the y values? I am trying to forecast energy consumption in buildings and that is dependent on seasonality and temperature. Can Prophet also make the predicted y values based on predicted temperature? If not, do you have any other recommendations to methods of prediction? Thanks!

[@NicholasRenotte]

Yup! It supports multivariate modelling.

[@sanaarafique]

Hi Nicholas, I have a training dataset and I'm trying to forecast for the following 7 days (after the last day in the training dataset) but my output shows a few days missing. How can I resolve the issue?

[@NicholasRenotte]

Heya @Sanaa, let me double check, so the forecast is missing days or you're getting errors when you try to forecast because days are missing in the input data?

[@sanaarafique]

The forecast is missing days and I’m not sure why.

[@NicholasRenotte]

@@sanaarafique can you impute the days? Possibly apply a mean or median durin preprocessing. e.g. www.kaggle.com/kmkarakaya/missing-data-and-time-series-prediction-by-prophet

[@joshsmart9178]

is there a way to increase the size of the graphs?

[@NicholasRenotte]

Heya @Josh, you can use plt.figure to reset the plot size: matplotlib.org/stable/api/_as_gen/matplotlib.pyplot.figure.html

[@jascbatalla]

awesome video!!! I just have couple of doubts: 1 how can we measure the error? like in linear regression? 2, How should we work with dates, say I want to forecast from July to December, do I need previous year data on those dates? is there a blank space of data I should leve in order to forecast?? If any one has more resources about working with time series I would really appreciate the help!! thanks a lot!!!

[@BB-ko3fh]

How was the model able to determine the daily seasonality when in fact you did not pass any intra-day (minute) data?! Really good video walkthrough; Keep up the good work!

[@NicholasRenotte]

Heya @B B, I took at look at this afterwards and realised that in fact we didn't have minute data. So you're right, it wouldn't be able to pick up daily seasonality! If we had more granular data it would though. Good pick up!

[@taufiqulhaque4987]

Amazing Nicholas... Well Explained, No complexity, well production. Would you please create another time series forecast model, where we can predict sales or stock prices for future (inputted) dates and times?

[@NicholasRenotte]

In the pipeline! Got some more stock/finance stuff coming soon :)

[@harikrishnangp959]

kzbin.info/www/bejne/ZnbCZGSNprutebs&ab_channel=PythonEngineer

[@mahmoudsamir9537]

Very good explanation, thank you a lot.

[@charusamaddar6550]

Hi! This is a great video, I enjoyed the quick way of forecasting so easily. But as soon as I tried to install the fbprophet package. I ran into error. Command errored out with exit status 1. I am windows, with anaconda jupyter notebook having python 3.9 Any tips on installing it successfully ?? Thanks!!

[@NicholasRenotte]

Heya @Charu, was there a more detailed error?

[@charusamaddar6550]

@@NicholasRenotte Thanks for responding. I got it resolved using this solution. hemantjain.medium.com/solution-for-the-error-while-installing-prophet-library-on-windows-machine-d1cc84adbafc And Also I had to disconnect from any kind of VPN.

[@NicholasRenotte]

@@charusamaddar6550 ahhhh got it! Awesome work and thanks for sharing!

[@luisfelipeperezherrera5887]

hi i am having trouble installing fbprophet on my pc could someone help me with this

[@alvaroflores453]

Goodie, just curious on how it generated a "within the day" plot without that info, but seemed to pick up some consistent trend haha. Maybe those are the priors showing as it looks quite symmetric

[@telander1484]

Great video! Just one question; how is hourly seasonality available when you have not specified any hours on the dataset? The data seems to be total sales/day for a single product in a single location.

[@telander1484]

Nevermind, just saw the comment by B B. Still interesting that it tries to produce hourly seasonality!

[@telander1484]

I'm going to predict incoming chats and calls/hour for my company's customer support schedule

[@NicholasRenotte]

Awesome use case! I thought it would have thrown up some additional errors when I was passing the data (tbh I shouldve been paying more attention as well!). How's it going so far?

[@telander1484]

@@NicholasRenotte Preparing a demo for my boss, I don’t have access to the real data yet! I acutally work as CS but i want to be data analyst!

[@NicholasRenotte]

@@telander1484 awesome stuff! Let me know how you go!

[@yashpatil9564]

Can we use prophet for multivariate forecasting . IF yes , can you make a tutorial on it

[@Sunsets_LoFi]

So detailed explanation

[@egegirsen]

You are the best I love you man

[@ferzim63]

Nice, but I still have problems installing pystan and fbprophet, how can this be so dificulkt, it has so many errors

[@sigmaakhil9990]

i got error in installing fbprophet -is 'pip install fbprophet ' is the command?

[@rangerxd1225]

I have same problem with installing. I used Anaconda prompt too. It didnt work.

[@sigmaakhil9990]

@@rangerxd1225 what will do to solve it?

[@rangerxd1225]

@@sigmaakhil9990 no solution yet for me. Somewhere i saw that you need to have python 3.7 for Fbprophet. It doesnt work with python 3.9 . Do check your version. And try to revert back

[@sigmaakhil9990]

@@rangerxd1225 my python version is 3.9..

[@sigmaakhil9990]

@@rangerxd1225 what's this gcc error

[@vjramyasaravanan2212]

best tutorial ever

[@srjoy101]

In my dataset , I've multiple products . Its like : Date product1 product2 date1 2 sales 4 sales date2 3 sales 6 sales How do i predicts all the products at once.

[@srjoy101]

I found this for multivariate situtations : model_new = Prophet() #instantiate Prophet model_new.add_regressor('humidity') model_new.add_regressor('wind_speed') model_new.add_regressor('meanpressure')

[@maherkarim693]

explained with such incredible simplicity. have you gone into more detail on seasonality into another video? keep up the good work!

[@NicholasRenotte]

Hi @Maher, thank you! I haven't but I can if it's a video you'd like to see?

[@diegobravoguerrero]

@@NicholasRenotte yes please! And thank you! I know how hard is to produce a single video. Great work on your channel.

[@NicholasRenotte]

@@diegobravoguerrero added to the list. Thanks so much!!

[@rowlandoshiotse9684]

very detailed, easy to understand, concepts were also explained. nice one Bro. can i use this to predict future football scores for my team?

[@lolhiphop6178]

Hey Nicholas, is the neuralprophet a kind of GAM? Can you still interpret it with the neural network from neuralprophet? what is the advantage of this neural network? thank you for coming answers :)

[@NicholasRenotte]

I don't believe so, under the hood it's using a Neural Network called AR-Net (github.com/ourownstory/neural_prophet). I'm still looking at what the performance bonuses are like versus something like regular Prophet.

[@lolhiphop6178]

@@NicholasRenotte thanks for your answer, that helped me a lot

[@NicholasRenotte]

@@lolhiphop6178 no problemo! You're most welcome!

[@chanhopark5506]

thanks a lot!! You are my lifesaver.

[@NicholasRenotte]

So glad you enjoyed it @Chanho!

[@the_yb7]

pls give access to the slides attached.

[@whiteboardmachinelearning7693]

please make a video on multivariate time series forecasting

[@spqri3]

The best, as always. Thank you!

[@Niteshkumarv]

can you tell me the code for future date that may be 2 -nov-2021

[@NicholasRenotte]

Depends on your end date, when you make the future dataframe just ensure that it extends out past 2 Nov 2021!

[@pooranigopu2502]

I'm facing Key Error: 'metric file'.

[@NicholasRenotte]

Where is the error occuring?

[@anaclaramedeiros4110]

Hello Nicholas, thank you so much for your explanation, it was very nice and clear in a often complex subject as Time Series...Do you have any recommendation in regard to a demand forecast for SKUs? They are phamaceutical products, around 6000 of them, each of them with a different ID. We are using prophet now, but some people are suggesting a LSTM model which to me seems to be very complicated. Also, we needed a model that could take into account exogenous variables that i am also not sure how to add into the model as a feature.

[@NicholasRenotte]

Hey Ana, I'm presenting on how to do that this week: online.datasciencedojo.com/events/sales-forecasting-python-prophet-2

[@Learner_123]

Hi Nicholas, Thanks for the, as usual, excellent tutorials. I have to prepare a forecasting model for nearly 50K unique products. I know it can be done by looping each product and forecasting separately, but this would generate as many models as the number of products which does not seem to be a good solution. Can you suggest how to approach this problem? Do you advise an algorithm other than Prophet, which can be helpful here? As can be seen in your tutorial, Prophet takes 'ds' and 'y' for training, can we add more input features to the algorithm?

[@henrystevens3993]

You can try Holt winters model

[@Learner_123]

@@henrystevens3993 Thanks, but the question is how to avoid a loop for training multiple items?