ARIMA in Python - Time Series Forecasting Part 2 - Datamites Data Science Projects

[@philwebb6671]

@20:53 you were right to begin with. Your test data starts at 27, so your predictions should start at 27. Also, the stop parameter is inclusive. 27 to 36 is 10 months. You should have stopped at 35 and calculated rmse on 9 pairs of data.

[@philwebb6671]

Then, @29:50 you moved the first data point of you test data back to 26, but didn't moved the last point of your training data. So, your predictions start at 27 and your test data starts at 26. That makes them overlap. If you had kept the indices they would plot correctly regardless of where you started. Your first graphs were correct. The predictions actually lag one month.

[@SatishKumar-ph6hb]

Amazing explanation of time series model development in Python..Thank you so much.

[@DataMites]

You're welcome

[@sathishd1419]

Very simply explained, before this I have gone through a number of materials for understanding ARIMA but had never got this clarity. Thank you so much.

[@DataMites]

Glad it helped!

[@michaelngugi158]

My first ever KZbin Channel to subscribe to. Easy to understand. Good Stuff Ashok keep them coming

[@DataMites]

Thank You!

[@gloriyathomas]

One of the best time series tutorial for beginners!! Thank you..

[@DataMites]

You're very welcome!

[@deegee9497]

Man, you saved me! Thank you so much, my final year presentation is in next week and I don't know what I would do without you. :)

[@DataMites]

You're Welcome

[@TheMunishk]

Ashok,although i will have to code first to fully grasp the concepts, but must tell that you explanation brings lot of clarity

[@DataMites]

Thank you!

[@gagangupta7840]

Both the videos on Time Series are wonderful

[@DataMites]

Thank you

[@anaghadamame196]

I learn something new today in very easy way👍 Thank you sir😊

[@DataMites]

Glad it helped.

[@pranavk1473]

Definitely , you guys deserve a lot of subscribers mannn !!!! Bang on explanation on both theory and practical ...

[@DataMites]

Thank You!

[@tho-mas-l]

Unfortunately, this get some of the basics wrong: if the time-series is not stationary, it doesn’t automatically follow that you need to difference it. Differencing only helps for stochastic trends, but not for linear time trends. In the latter case, which is quite common, you can simply make the series stationary by subtracting the trend (or in the case of an ARMA model by simply making time an exogenous regressor). Differencing only helps if the time-series is integrated (“random walk”), which you can confirm by performing the Dickey-Fuller test. From the graph shown it's not possible to tell which of these cases we are dealing with.

[@DataMites]

Thanks for your comment, we are planning to give some study material to the students on different parametric and non-parametric statistical test for stationarity. In this video, we tried to help the people getting started with time series and wanted them to be aware of stationarity before going forward in time series prediction.

[@Zero-gp5vl]

btw can u give some advices about modeling time series where data was distributed by hours ?

[@HaiderAli-pp9go]

@Thomas Loeber Sir would you be kind enough to recommend/link me to some resources concerning time series forecasting. I am quite new to the field but finding it difficult to find good literature and resource. Actually my thesis is related to wind energy forecasting and I believe and introduction to time series forecasting will be a good stepping stone. Any help, insight and guidance will be very much appreciated.

[@kasunathapatthu2372]

​@@HaiderAli-pp9go bro did you find a relevant video or source to learn about wind energy forecasting? Also my thesis is about wind energy forecasting.

[@HaiderAli-pp9go]

@@kasunathapatthu2372 No brother unfortunately I am still struggling. I have found something but it is not really working. If you want, we can have a teams meeting so that we can discuss what each of us want to achieve and how we can help each other?

[@devchuvak1716]

It was really good explanations! Thank you for your tutorial.

[@DataMites]

Glad it was helpful!

[@SANJUKUMARI-vr5nz]

This coding made me confident

[@DataMites]

Glad it helped

[@jitendrarathod2455]

18:55 I think there should be range (start=27,end=35)..then will get 9 over there..

[@DataMites]

"Hi jitendra rathod, Yes you are correct. Thanks for pointing that out."

[@andersonaraujo1992]

Thanks a lot.. I learned a lot in two videos. Now I am gong to try some projects to train.

[@DataMites]

Great!

[@waisyousofi9139]

thanks for clear expln of p q and d. no vid seen like this to expln p and q and d.

[@DataMites]

You are most welcome

[@venkateshakhouri9541]

I have one question When we have removed the stationarity using diff() function, why are we using the sales dataframe for the model instead of sales_diff

[@tollpatsch5969]

My thought about that is, that we use the diff for our own visualization. With Arima and the parameter d the data get differentiated.

[@venkateshakhouri9541]

@@tollpatsch5969 so we we differentiate just to get the values of p,d,and q?

[@DataMites]

sales_diff is for explaining the concept of making data stationary. In actually model we use original data and 'd' parameter in ARIMA hold the order of the differencing. We don't need to explicitly do the differencing when using ARIMA.

[@manikmahashabde2946]

Thanks a lot Sir. This is the best explanation I have come across. Subscribed!!!

[@DataMites]

Thanks and welcome

[@tioputrasalis9807]

thank you for your knowledge, may i know what kind editor you use for python?

[@DataMites]

"Hi Tio Putra Salis, If you want notebook like interface as used in video, you can go with jupyter notebook. You can use ide like pycharm professional edition that is paid service or you can go with open sourced editor like vscode, atom or pycharm community edition."

[@estebanarroyo8159]

Why do you use the sales time series (non-stationary) instead of sales_diff time series (stationary)?

[@DataMites]

Esteban, We should use stationery and this is taken care in ARIMA model with d parameter from (p,d,q). d=1 means first order difference data.

[@shital6611]

@@DataMites Thanks for the explanation

[@sonaligrover9497]

@@DataMites - Yes, this is true for ARIMA. But for AR, we have again used non-stationary(sales) instead of stationary(sales_diff). Is this correct? Please respond, it'll be really helpful.

[@filizcamuz8119]

​@@sonaligrover9497 You have to pass the stationary data to AR Model. It does not handle non-stationary data, unlike ARIMA. So use sales_diff series for that model.

[@sambitnath9853]

Excellent explanation 👌👌

[@DataMites]

Thank you 🙂

[@laxmitalawar6604]

Awesome! Amazing explanation thank you so much sir..

[@DataMites]

Thank and welcome!

[@jaydeepdedaniya5493]

use (p,d,q) = (8,2,2) # it gives best fit among the all combinations from 0 to 10

[@PraveenKumar-pd9sx]

But how

[@DataMites]

Hi, you can certainly try those combinations.

[@kmnm9463]

Hi Ashok, Very nicely conceptualized and presented session on AR and ARIMA. I have a basic doubt - The original sales data is non-stationary and we need to transform it into stationary to make predictions. From steps: X = sales.values train = X[0:27] test = X[27:] model_ar = AR(train) model_ar_fit = model_ar.fit() What I think we are doing is creating an AR variable for passing training dataset for prediction. But the 'train' data is from the original non-stationary data so how does the model created be able to predict on the test data. I thought the sales_diff should be the dataset to slice and dice into train and test to pass on to the model for prediction. Let me know if I misunderstood something . Regards KM

[@sayyadsalman9132]

You are correct, I also face the same problem.

[@DataMites]

Sales_diff is for explaining the concept of making data stationary.We expect you to include code of sales_diff to see the prediction and compare with the original. In ARIMA model we use original data and 'd' parameter hold the order of the differencing. We don't need to explicitly do the differencing when using ARIMA

[@shazzadislam4905]

Thanks a lot for your very helpful tutorial...

[@DataMites]

Thank you!

[@SaurabhKumar-ic7nt]

Best explanation.........

[@DataMites]

Thanks you

[@devyaninitturkar8662]

Nice explanation ...thank u so much sir

[@DataMites]

Thanks and Welcome!

[@manishchabor]

Thank you, very nicely explained.

[@DataMites]

You are welcome!

[@sayyadsalman9132]

In the AR model, you did shifting and differencing to make the data stationary but, you actually used non-stationary data with AR model. X=Sales.values and then X[0:38] for training. Can you explain it?

[@DataMites]

That is done to show how non stationary data can bring mispredictions. We expect you to build Solution with difference data only.

[@layeeqshaikh3384]

hi, this is really simple, we would want a tutorial on VAR wherein there are multiple dependent variables, kindly do a video soon with a sophisticated dataset

[@mignoncharly]

Hi, whenever you got somethig like that. Thks for sharing :)

[@DataMites]

Sure! thank you for watching

[@PankajKumar-eg9sw]

Thank you explanation of AR and ARIMA model in lucid way, can you please make a video on Exponential smoothing

[@DataMites]

Sure

[@sgrouge]

Simple and clear.

[@DataMites]

Thank you!

[@rvind8285]

Do we have videos on Multivariate time series analysis?

[@rohan1427]

I am recently working a dataset like that, it is a PhD level problem but fun.

[@gregs138]

Great video, TY!

[@DataMites]

Glad you enjoyed it!

[@PavanKumar-rw3br]

Is stationary a pre-requisite to develop any timeseries model or is it a pre-requisite only for an ARIMA model?

[@DataMites]

Stationery is prerequisite for time series. In ARIMA model this is automatically taken care with d parameter so for ARIMA stationery is not prerequisite

[@javariailyas2429]

And also can we do all these steps in pycharm IDE

[@aleksandramazurek1364]

Thank you! This was amazing, very helpful and well explained :)

[@DataMites]

Glad it helped!

[@satyamsalokhe2996]

helpful, good explanation

[@DataMites]

Glad it was helpful!

[@oktofenno9622]

Amazing explanation of time series model development in Python..Thank you so much. i have a question: what if i,ve already got stationary data from the first. is it necessary to take df.diff() again . or we can skip that phase and later to put d value by 0 ?

[@DataMites]

"Hi Okto Fenno, If your data is stationary, there is no need to make it stationary by giving d value."

[@rohan1427]

Something i realised is that, even after hyperparameter tuning we have to fine tune our model then there's no point of hyperparameter tuning. I think it says a lot of about metric like AIC because it is not helping us get the best hyperparameter, maybe we should use MAPE during hyperparameter tuning as well and then take the best pdq and train our ARIMA. Just a thought !

[@DataMites]

"Here, there are lots of metrics that you need to go for while training. Here, lower AIC generally means lower RMSE but sometime you might get higher rmse for lower AIC. So there are lots of things that need to be taken into consideration. Every metrics has its significance and you can make the best decision once you know different metrics in and out. We appreciated your comment though."

[@Zero-gp5vl]

btw can u give some advices about modeling time series where data was distributed by hours ? I need some recomendations

[@theoreticalcomputerscienti4112]

How do we un-difference forecasted values of a differenced series in python to make the forecasted values and the test values in the same units? Thanks.

[@DataMites]

"Hi Scientist and thanks for reaching out regarding your doubt. While passing the values in ARIMA like here you are not passing the subtracted value, as ARIMA does itself that from d value in its parameter. So your output will be in the same unit as the test."

[@hariprasadv166]

at 2:28, as you have converted year and month with parser...but in my dataset i need to read it as ISO year and ISO week number....could anyone please help me out with this. ..?

[@DataMites]

"Hi Hariprasad, It would be easier if you had also put some rows of those date in your query so it will be easier to answer that."

[@MrTheothegreek]

hello, Great video! thanks, one question though.. Why is my print(model_arima_fit.aic) prints garbage instead of just the value? I am programming this in pycharm btw

[@DataMites]

model.aic will print AIC value in any platform, ofcourse. Could you share the full code so that I can find the reason for this?

[@aniketmlk6]

God bless you!!!

[@DataMites]

Thank you

[@rudzanimulaudzi7947]

Great video! Quick question, why dont you run the ARIMA model on the differenced data set why do you use the original one? or do you use the differenced one just to know the value of d? which doesn't make sense given that you use grid search to find the right pdq values. Hopefully my question is clear.

[@DataMites]

This was just to show however when you practising please use differenced data.

[@anilkalai4778]

During AR MODEL. why you didn’t take dataframe of sales_dif?

[@bouraimakouanda8737]

I do have the same question.

[@DataMites]

"Hi Anil Kalai, d value in parameter (p,d,q) stands for the same difference, so no need to do that."

[@wimavlogs6826]

Can we do time-series predictions by combining two models ( ARIMA + Nural Networks )? if it is, can you please do it with the video?

[@DataMites]

You can use LSTM for time series to get the best predictions.

[@adrijenie4105]

Is there is any maximum on p (on p,d,q)? or we could just choose anything as long as AIC minimized?

[@aashikasharma1616]

Yes, just keep minimizing AIC

[@DataMites]

Normally we go with value 0 to 5 as they are found to give you minimized AIC. But you can go beyond that range to some extent to get the better combination of p,d,q.

[@ramneetsingh2556]

Great Tutorial! Thanks

[@DataMites]

You're Welcome!

[@smitmalik6784]

i am getting an import error and Traceback while using from statsmodels.graphics.tsamodels import plot_acf plot_acf(dataset_name) any solution?

[@DataMites]

Please Check www.statsmodels.org/stable/generated/statsmodels.graphics.tsaplots.plot_acf.html

[@AmoghSapre]

Hello Sir, Can you pls guide how we can write the program to find the pdq combination for minimum AIC value within the logic you provided at video 35:12? so we can increase the range from 0,5 to 0,10 and don't have to look for minimum AIC value manually. Thanks and very much appreciate you trainings.

[@DataMites]

"Hi Amogh Sapre, and thanks for reaching out with your queries. you can tune parameters p,d and q by using three for loop one inside another like as follows: for p in range(your_range1): for d in range(your_range2): for q in range(your_range3): # Here you will get all the combination of p,d,q as said in part 1 of this tutorial and use this information to find minimum aic"

[@ibrahimrashid]

Check out AnticiPy which is an open-source tool for forecasting using Python and developed by Sky. The goal of AnticiPy is to provide reliable forecasts for a variety of time series data, while requiring minimal user effort. AnticiPy can handle trend as well as multiple seasonality components, such as weekly or yearly seasonality. There is built-in support for holiday calendars, and a framework for users to define their own event calendars. The tool is tolerant to data with gaps and null values, and there is an option to detect outliers and exclude them from the analysis. Ease of use has been one of our design priorities. A user with no statistical background can generate a working forecast with a single line of code, using the default settings. The tool automatically selects the best fit from a list of candidate models, and detects seasonality components from the data. Advanced users can tune this list of models or even add custom model components, for scenarios that require it. There are also tools to automatically generate interactive plots of the forecasts (again, with a single line of code), which can be run on a Jupyter notebook, or exported as .html or .png files. Check it out here: pypi.org/project/anticipy/

[@faizibrahim7646]

Thank you for sharing this vast knowledge with us. I need your help please, after fitting the ARIMA model and making the forecast, I want to calculate the residuals of the model and use a machine learning algorithm to model the residuals. I am having issues on how to calculate the residuals, can you please help me out with a simple formula for calculating the residuals or any guide that can help. Thank you

[@DataMites]

Calculate the residuals by subtracting the predicted values from the actual values. Residuals represent the differences between the observed data and the predictions made by the model.

[@anuragdeepak9975]

Sir, do we have to check for linearity of dataset before applying arima and how can we do that. Also, do we have to take stationary or original dataset for finding the values of pdq. Arimax lecture through python should also be made.

[@DataMites]

"You need to take stationary dataset to find pdq. Thanks for the suggestion, we are constantly publishing new videos regarding different new and improved approach. Keep checking our channel and subscribe to get the notification."

[@karankapoor6624]

Hi great explaination , one question though can the AIC value be negative ?

[@DataMites]

Hi, Thank you. Yes, the AIC score can be negative.

[@frankconte2457]

Thank you for a wonderfully understandable video. However, the "mean_squared_error(test,predictions)" threw an error. The error was as follows: "ValueError: Found input variables with inconsistent numbers of samples: [9,10]. What am I doing wrong?

[@DataMites]

There is mismatch count of your x and y

[@frankconte2457]

@@DataMites Please elaborate

[@DataMites]

@@frankconte2457 Please check your x and y splits

[@ragini2669]

love you man, you saved me today

[@aashikasharma1616]

Yeah, same here

[@DataMites]

Thank you!

[@jiajiaou4293]

Great video! helped me a lot. Thank you for sharing:)

[@DataMites]

You're welcome!

[@Adinasa2]

this is a very good video, you could have added seasonal_decompose and adfuller test also in the video!!

[@DataMites]

Thank you for the suggestion!

[@Adinasa2]

@@DataMites i have implemented it in my jupyter notebook. Please drop me an email if you want me to share my notebook.

[@DataMites]

Will do it in the future session.

[@haneulkim4902]

Thanks for your video! So model with lowest AIC outputs best prediction? It didn't look like it in your video :(

[@DataMites]

The AIC function is 2K - 2(log-likelihood). Lower AIC values indicate a better-fit model, and a model with a delta-AIC (the difference between the two AIC values being compared) of more than -2 is considered significantly better than the model it is being compared to.

[@sandile4764]

very good! :) Thank you

[@DataMites]

You're Welcome

[@meilingjin5164]

hi , this is the best tutorial I ever see for time series prediction thanks, I have one question, the plot_acf you made, why it started at 0 but the value is still 1?

[@roy6378]

I'm guessing that at 0, you are comparing the series with itself so the values will be the same, hence an auto-correlation of 1?

[@DataMites]

"Hi Meiling Jin, those are the lags and in any programming, we start the count from zero. For more check this out: www.statsmodels.org/stable/generated/statsmodels.graphics.tsaplots.plot_acf.html"

[@alimahmood4158]

This is extraordinary kindly please tell that what are the drawbacks of AR that does not simplify our problem so we Use arima

[@DataMites]

Thank you very much. We will do it

[@jjanime6130]

Hi. Have one question, if I try write in pdq 9 as p, I got mistake about ar coeff. are not stationary. How can I add seasonal pdq

[@DataMites]

"Hi Валерий Куринный, thanks for reaching us regarding your queires. But can you please elaborate your question?"

[@hannesbreitfeld5315]

i used my own data and tried to build my own model but when executing "predictions = model_arima_fit.forecast(steps=104)" like in 28:00 of your video i simply get an empty array. Could you imagine why?

[@hannesbreitfeld5315]

As some of you guys may facing the same issue the solution was pretty easy: I just had to restart the kernel xD

[@DataMites]

Try to keep steps 1 and check prediction.Similarly check for 2, 3, and 4.If not check the periods on you are working.

[@maheshreddynimmala5795]

Sir I have one dought why we split the data into train and test is there any reason

[@chetan923]

I'm not the Sir here, but seeing your question past 3 weeks, I would like to help you understand. of course there is a strong reason why we're splitting the data. this is called machine learning, in the test data you're teaching the machine how the data moves with time and then, you're making it predict the next data similar to test data so you can match and check how efficient the machine is in predicting any new data introduced to it.

[@DataMites]

If you build a model on the entire dataset then evaluation of the model is not possible. For evaluation purpose, we hide some data from the trained model and test its performance on the test.

[@WafiKhanjerMusic]

im getting negative numbers, what is wrong and would be the reason for that?

[@DataMites]

Check our dataset again. And tally it from video

[@ammarsulaiman4555]

thank you so much !!! but why we did not work on data after we made diff (stationary data ) in AR model !!!

[@DataMites]

That was to show you however while you code please use the diff one

[@pushpag1076]

good afternoon sir, I have error, model_arima=ARIMA(train,order=(1, 2, 1)) ,i have run this command but its come error Error:TypeError: Cannot cast ufunc subtract output from dtype('float64') to dtype('int64') with casting rule 'same_kind'

[@pushpag1076]

sorry sir ,i got the output .now only i saw the another person command,thank u sir very useful for your video

[@DataMites]

@@pushpag1076 You're welcome

[@harjotsinghsaini1085]

Thank you very much. Really helpful

[@DataMites]

Thank you!

[@ArunjyotiNayakchinu]

Good explanation sir. I have dataset of toll plaza (6 to 8months) and have to predict for next 3 months.there u have to get number of all types of vehicles and aslo how much revenue we are getting in that 3 months. How to do that by which method?

[@DataMites]

"Hi Arunjyoti, thanks for reaching us in the comment section. If there is enough pattern that your any time series algorithm needs you can go with 8 months data, but to use it to generate the next three months data might be a little problematic as your model will have more errors every month you go forward. If your dataset target variable depends on only one variable and to itself (i.e univariate prediction), you can go with arima and if you need to consider multivariate data (i.e you need to take multiple variables into consideration before predicting your target) it's better to go with lstm and deep learning."

[@sendrapyansyah2993]

Hello, if i use AUTO ARIMA function, after differencing the data to make stationery, is that ok???

[@DataMites]

Differencing is not required for ARIMA as the model itself will do it.

[@anasitanggang]

tq sir

[@DataMites]

Welcome

[@chandanjha3205]

I liked the tutorial a lot but I am not a fan of trying out various values of p and q. You earlier showed that d will be one because 'differencing' it once removed the trend so why later we are using d as 2. And howcome q can take different values. Lags should be somewhat accurate. Isn't it.

[@aashikasharma1616]

Because d=1 doesn't guarantee that d=2 can't be a good fit. So we have to try different values.

[@DataMites]

"Hi Chandan Jha, thanks for your comment. In the earlier portion, we used one differencing and it was moving our dataset closer to stationarity and stationarity is a relative term. Later on, we found out that two times differencing will move us closer to stationarity than one time differencing. And regarding different q values, can you please elaborate?"

[@nomanshaikhali3355]

Hey, I want to predict buysell signals of stocks for next 30 days from now! I have done with your video but the problem is buysell signals col contain(0 and 1) values but forecasting gives in points or continuous values b/w 0 and 1

[@DataMites]

You can use LSTM for such cases.

[@nomanshaikhali3355]

@@DataMites No dear, I don't want to use Deep learning model!!

[@DataMites]

@@nomanshaikhali3355 Based on your query alone, we could suggest that you can put some threshold, let's say 0.5 if it is greater than 0.5. change it to 1 otherwise 0.

[@chillwithme798]

why do you use data transform by Diff?

[@DataMites]

Hi Công Vinh Trần, Differencing is a method of transforming non-stationary data to stationary and time-series models work under the assumption that the underlying data is stationary, i.e the mean, variance, and covariance are not time-dependent.

[@kesotics8503]

thank you very much sir

[@DataMites]

You're welcome

[@ckvaram]

Good video. I have one question at 10.54 min, Could you please explain If acf plot is quickly decaying positing and negative then how can we say data is stationary ?

[@DataMites]

ACF helps to know the correlation between lags. If there is a correlation between lags, then data is non-stationary otherwise it is stationary. Therefore, if the ACF plot is decaying then there is no correlation between lags.

[@sunilnarwaria7452]

nice explanation

[@DataMites]

Thank you

[@kevinalejandro3121]

If I only want to consider a certain lag in my ARIMA model?? for example only consider the lag 3 but i don't want the lag 1 and 2 in my model, How can i do that ??

[@DataMites]

Take p=3

[@judesavio1168]

Is there any restrictions on the 3 parameter values (p,d,q) ? my doubt arose as to why you specifically selected the range as [0 : 5] Is there a specific range or how to find the correct range of values for p , d ,q without trial and error ? because if the correct value of p is hundred finding it out manually will be a huge burden. Please share some insight

[@LUMIGOCHA]

I'm not expert in data analysis, but i've read that having many parameters in a model can create an overfitting. With ARIMA(4,2,4) for example, you would have 10 parameters to calibrate and each one of them will have an error asociated. So you need to be carefull of what you are aiming with your model. Hope these articles give you a better inside: statisticsbyjim.com/regression/overfitting-regression-models/ …… stats.stackexchange.com/questions/17565/choosing-the-best-model-from-among-different-best-models ….. www.ejwagenmakers.com/inpress/VandekerckhoveEtAlinpress.pdf ...

[@DataMites]

"No there is no strict rule to choose it, its one of the best practice to use that range and it was found to be the optimal range to get the optimal result from ARIMA. We are not aware of any of those methods without trial and error to find the best p,d,q value. You need to tune it and it can be done by evaluating the current metrics from the previous one."

[@MyChidananda]

Great tutorial. But I just didn't the indexing for test data set, why it is X[26:] as per my understanding it should be X[27:] and 9 further data points to be predicted. Please explain me this part.

[@DataMites]

"Hi Chidananda, and thanks for reaching out regarding your doubt. For training its X[:27] to get value from index 0 to 26 and for test its X[27:] index 27 to 35."

[@devdaskamath975]

what is the difference between d and q? d takes the difference, also q which is the error is obtained by taking the difference? so what is the difference between d and q?

[@DataMites]

p: The number of lag observations included in the model, also called the lag order. d: The number of times that the raw observations are differenced, also called the degree of differencing. q: The size of the moving average window, also called the order of moving average

[@smallgirle]

I have an error when i try to fit the ARIMA model saying :"Cannot cast ufunc subtract output from dtype('float64') to dtype('int64') with casting rule 'same_kind' " even when i use the exact code mentioned above. Pls Help

[@akshaygoel7025]

use this model_arima = ARIMA(train.astype('float32'),order = (4,2,1) ) might resolve your problem

[@DataMites]

"Hi Hunsii Ashar, thanks for reaching us regarding your queries. From the error statement, we will suggest you typecast all your value to float before using ARIMA."

[@ashishtinker8119]

Sir, which software is this. I have opened pandas. I haven't found this, Kindly help me please. I need to forcast banknifty please

[@DataMites]

Hi, not clear with your query. Kindly elaborate. If you want to code, you can use Jupyter notebook. Or if you are referring to the Time Series Forecasting model, it is a method and not a tool. you need to import it from the corresponding library.

[@stavroschanidis919]

You 've made a mistake at the searching of AIC. If you see clearly the minimum value of AIC is 290,.... and it corresponds to the combination of (1,2,3). Just for the shake of right evaluation, otherwise your tutorial is really great. Thank you!

[@metalflames4]

How did you know?

[@stavroschanidis919]

@@metalflames4 If you pause the video at 38:20 exactly, you will see right in front of you in the middle, the result of 290,..... that corresponds to the (1,2,3) combination of pdq

[@DataMites]

"Hi Stavros, thanks for your comment. In the range of 0 to 5, that minimum AIC was 290, and that is correct. But there are other metrics to get into consideration too before finalizing the and we have gone through this (not in this video) after that we found the optimal value of p,d,q to be the value used at last. AIC value is one of the ways to tune your p,d,q value. We will make more videos regarding those metrics and make sure your keep on checking our channel so that you will not miss that."

[@RahulDas-ki7tg]

how to predict future lets say next 6 months sales value??

[@mcchandrashekar]

If you have sales data till this month, you can predict next 6 months sales figure in the same way has Ashok sir explained here. The only the difference is, you can't verify the prediction you got is accurate or not until you get the actual sales figure in future. :)

[@riyazbagban9190]

iin ar model periods = 1 can we take 2 or 3 please explain what is 1, 2 ,3 thank you

[@DataMites]

Yes, you can. In this model the current value depends linearly on the previous or past terms. Means the previous terms are the predictors here. So when periods=2, it takes 2 preceding values for predicting the next value.

[@laizerLL572]

thanks 👍 sir for your helpful tutorial to me can you share your notes for reference

[@koushiksrinivas2459]

sir, you told you made the series stationary but the stationary series is sales_diff but you used sales values for modelling. Can someone explain that part?

[@RecursiveDimension]

I didn't see the whole video, but rather bits and pieces of it. But I'm pretty sure he did manual differencing to show the "Integrated" portion of the ARIMA technique. Later he is actually using the full ARIMA model imported from StatsModel. So, the "d' is the differencing portion. There he actaully tries different p, q, & d values to minimize the AIC number. Hope that helps.

[@DataMites]

"Hi Koushik Srinivas, and thanks for reaching out regarding your doubt. While passing the values in ARIMA like here you are not passing the subtracted value, as ARIMA does itself that from d value in its parameter."

[@clocks7578]

How can you identify the parameters (p,q,d) of ARIMA model?

[@DataMites]

"Hi Clocks, and thanks for reaching out with your queries. At basics, you can tune parameters p,d and q by using three for loop one inside another like as follows: for p in range(your_range1): for d in range(your_range2): for q in range(your_range3): # Here you will get all the combination of p,d,q as said in part 1 of this tutorial and use this information to find minimum aic In tutorial itself, itertools package has been used for it."

[@mayanktripathi4u]

Getting below error. Though followed the same steps. Does any one had same issue, if so how did you resolved it? Please share. TypeError: Cannot cast ufunc subtract output from dtype('float64') to dtype('int64') with casting rule 'same_kind'

[@DataMites]

"Hi Mayank, thanks for reaching us regarding your queries. From the error statement, we will suggest you typecast all your value to float before using ARIMA."

[@UserRoot15015]

i am working on a time series project as in intern and i got a negative variance! what does it means? how can i solve it?

[@DataMites]

Please subscribe yourself on tribe.datamites.com/

[@asmitadadhich3630]

Very nice video :)

[@DataMites]

Glad that it's helpful.

[@arvindnaidu7471]

can we use pd.to_datetime( ) instead of paser method

[@DataMites]

Yes. But you need to reassign to the column. Parser is easy and straight forward.

[@DataMites]

yes. but parser is simpler and straight forward.

[@GalacticEarthChronicles]

Sir how I set the data for forecast for next 12 month from last date

[@DataMites]

Hi you need to look for start and end parameer in ARIMA.

[@sreeramsaravanan8132]

Explain about simple exponential smoothing

[@DataMites]

Single Exponential Smoothing, SES for short, also called Simple Exponential Smoothing, is a time series forecasting method for univariate data without a trend or seasonality. It requires a single parameter, called alpha (a), also called the smoothing factor or smoothing coefficient. This parameter controls the rate at which the influence of the observations at prior time steps decay exponentially. Alpha is often set to a value between 0 and 1. Large values mean that the model pays attention mainly to the most recent past observations, whereas smaller values mean more of the history is taken into account when making a prediction. from statsmodels.tsa.holtwinters import SimpleExpSmoothing # prepare data data = ... # create class model = SimpleExpSmoothing(data) # fit model model_fit = model.fit(...) # make prediction yhat = model_fit.predict(...)

[@pushpag1076]

i have tried pdq in for loop ,but there is no data generated, please help me sir

[@DataMites]

"Hi Pushpa G, and thanks for reaching out with your queries. If you are having a problem with the given approach from itertools, you can tune parameters p,d and q by using three for loop one inside another like as follows: for p in range(your_range1): for d in range(your_range2): for q in range(your_range3): # Here you will get all the combination of p,d,q as said in part 1 of this tutorial and use this information to find minimum aic"

[@ashirwadparasar4752]

why difference dataframe was created when only the original dataframe was used for modelling, forecast and predictions?

[@DataMites]

We need to see how the model works with both the data. Here it is done with the original data frame. You have to try with a different data frame. And you can check the model's performance for both the data.