Tutorial 33- P Value,T test, Correlation Implementation with Python- Hypothesis Testing

[@srinathyasoda5545]

Great Video, Kudos! But, we should never say "we are accepting Null Hypothesis", we SHOULD Say it as "We are Failing to Reject Null Hypothesis", as there is always a possibility of error that lies with our sample data

[@shreyaskulkarni7612]

It's really true

[@ammar46]

Height and age doesn't follow Poisson distribution, it follows normal distribution.

[@zijieliu6654]

This is the best explanation of the T test on KZbin no doubt!!!

[@nilupulperera]

Thank you Krish for the introduction of statistical tools in python. Now only I realized how to do comprehensive statistical analysis without depending on Microsoft Excel (and addon software apps) which has limited capabilities in the Data Science field.

[@manikantasai721]

Very near to...100k sir congrats sir!

[@ivancarrillo1889]

Thanks for the explanation. By the way this is the second video from you I watched. I understood you much more clearly this time I guess because of the microphone (non native English speaker). Please keep it in mind.

[@mallikharjunv6805]

Thank you so much Krish ..excellent.

[@Ghodkeshubham6cool]

Very great learnings✌🏻

[@shrikantagrawal6642]

Summary: One categorical variable One sample t-test Two categorical variable Chi square test One continuous variable T test Two or more continuous Correlation and then T-test variable One Continuous and one T test categorical which has two categories One Continuous and one ANNOVA Test or more categorical which has more than two categories Two variables and you want Two sample T test to compute if their means are different One variable and we have Paired sample T test created one more variable based on first variable by adding some proportion to it on time basis @Krish - Please suggest

[@cinemascope8847]

Shrikant Agrawal please answer this Krish as it makes sense for all of us

[@narensingh6728]

Thanks bro

[@SameerAli-nm8xn]

You are great Sir, that's a lot.

[@mommysboy8015]

Great teaching skill

[@pavanaramu721]

The previous video teaching was excellent. Present video need to include chalk board activity for better understanding......

[@sumittagadiya3497]

very good explanation sir, thanks a lot

[@venkivtz9961]

Hi Krish, your explanation and example is excellent. But a small correction to the conclusions at the end of the test. We have to conclude the statement with respect to the alternate hypothesis. We should never say that "we are accepting null hypothesis".

[@sandipansarkar9211]

watched the video for the second time and practices on to Jupiter notebook.Thanks

[@harishkumar-zx6vg]

Y he used np. Random. Seed (6)?

[@harshdewangan1951]

@@harishkumar-zx6vg np.random.seed(n) is used to make the random number predictable, i.e., we will get same set of numbers whenever the code executed

[@skumarr53]

Thank you so much, sir, for the effort you are putting to educate us all. I want to make a career transition in Artificial Intelligence in computer vision and the NLP processing field. My question is do I need to be familiar with the ML concepts like feature engineering etc and ML algorithms or is it enough if I focus only on Deep Learning. I don't see much overlap between those two but both are treated as part of Data Science in the industrial setup.

[@snehakoul9818]

0

[@adiflorense1477]

1:49 krish, i have question. what t-test we use to see difference in machine learning model?

[@soujanyabagam2034]

sir, what is the difference between mean that you are passing as an argument in possion distribution and mean you are calculating subsequently?

[@shivangikdesai]

+1

[@001Debjeet]

btw congo on 100k silver incoming

[@sandipansarkar9211]

Awesome video Krish but don't forget to practice on Jupiter notebook.Thanks

[@batulkhan9772]

HEY, @ 7:07 minutes you're telling to reject the null, but the p_value is more than the 0.05

[@swarajkumarsahoo4736]

yeah, and the output shown is "We are accepting null Hypothesis"

[@questforprogramming]

We fail to reject the null hypothesis, because it is > 5%. 74% >5%

[@questforprogramming]

@@swarajkumarsahoo4736 he didn't run that cell at all. He ran previously means not the n video. So that output is wrong I guess and what he said is also wrong

[@BlueSkyGoldSun]

Yes me to iam confused , did he make a mistake?

[@06madhav]

Bhai, incredibly clear video. One doubt- how to go ahead with the hypothesis testing looking at the dataset? Means, how to decide whether any sort of such tests are required to be done on the dataset?

[@mishuchugh1777]

@madhav srimohan..did u got this?

[@divyanshuaswal1843]

@@mishuchugh1777 did u got this?

[@wealth_developer_researcher]

Sir, I have a doubt. On timeline 7:03 you said we reject null hypothesis in this case. But p_value > 0.05 and output is we are accepting null hypothesis. Please correct me if i am wrong

[@kaifahmed316]

Same here

[@BlueSkyGoldSun]

Iam also confused

[@sh__--]

Please correct guys Accepting the Alternative hypothesis and rejecting the null hypothesis is the correct answer. Mistakes will happen sometimes😊 I am also a learner..👍

[@louerleseigneur4532]

Thanks Krish

[@rambaldotra2221]

Thanks a lot Sir

[@Himanshusingh-ep1hc]

@10:35 the p value 1.1390 which is greater than 0.05 but still its printed rejecting null hypothesis ?

[@Abhishek-st4mu]

same here, @10.00 i confusing on that statement, how can 0.05 is greater than 1.139

[@sumitmaiti2218]

Great explanation sir... It clears the understanding of the concepts... I have just one doubt: How are we selecting which statement to be the Null Hypothesis and which one for the Alternate Hypothesis? Because based on that and the p value, we would come to the conclusion.... Thanks :)

[@samerrkhann]

Usually null hypothesis is used when we say there's no difference between two groups. For example, you draw a sample from a population and want to check if there is any difference between the mean of the sample or mean of the population. You will make null hypothesis that the two means are no different. Similarly when comparing two groups if you want to check if there means are same, you will develop null hypothesis that there are no difference between two. One last example, first let's say you flip a coin 5 times and get heads 5 times. You will make a null hypothesis that my coin is no different than the normal coin. Hope this helps :)

[@snehalpophale6287]

Thank you so much!

[@pratikshagwalwanshi8676]

When we already established poisson mean (mu) as 30 in classA_ages=stats.poisson.rvs(loc=18,mu=30,size=60) Then why do we get different value for classA_ages.mean()?

[@mahenderboda1339]

I also got the same doubt got any answer?

[@madhavilathamandaleeka5953]

I also ....☹️.....and how can we take those mu values ..?? Plz anyone clear my doubt

[@pratikshagwalwanshi8676]

Nope didn't get it yet. If someone gets this doubt clear then please tell.

[@sashpatra88]

Hi Krish, Can you please share the ANOVA Implementation with Python video as I couldn't find it in your list?

[@madhureddy5328]

Why we do P or T or anova test? If we come to conclusion what we do with the dataset

[@gopichand8874]

Have you got the answer ?

[@Arasu89]

Hi Krish, After rejecting the Null Hypothesis or accepting the null hypothesis, what is the next steps we will do with data. Do we remove the features?

[@anveshpoloju7331]

Hi Krish, for beginners can you please suggest 'order' of preparing for DATA SCIENCE... For example 1st statistics 2nd python 3rd ML 4th DL.... Or simultaneously. Where to start exactly is confusion for many people.... Thank you

[@megirija1897]

pls upload video on impementation for anova and chi square test...

[@somtonnamah5734]

please i would like to know if the distribution of data groups matter when checking correlation

[@sumitsaurav1710]

Hi Sir The value of mu selected as 30 in "classB_ages=stats.poisson.rvs(loc=18,mu=30,size=60)" is mean of what? and how does it differ from classB_ages.mean()

[@kiranchowdary8100]

ya same doubt i think mu is possion parameter

[@ankitgadwe2200]

@@kiranchowdary8100 You are right. It is some parameter. You can check it here: docs.scipy.org/doc/scipy-0.15.1/reference/generated/scipy.stats.poisson.html

[@architagarwal7379]

Bootstrapping is enough to implement this all t test, chi swuare test etc ??

[@payalbhattad8048]

Hey Krish, it was a great explanation. I have 1 doubt though, I am trying to use it on my dataset where I am finding if there is a significant difference between gender based on salary. But I am getting p-value as nan even if there is no null values in the dataset. Whereas if I am performing the same on SPSS it gives a p-value of 0.43. any suggestions?

[@soujanyabagam2034]

sir, are you using poisson distribution to create a fake data set of ages since we dont have a real data set?

[@veereshbk4394]

99k subscribers were there while making this video, now it is 262k subscribers, means 262k-99k people are on the way to become data scientist during covid pandemic. 2021 will have less demand for data scientists as supply is increasing! This is just sample testing

[@shahidraza7965]

Can someone tell me why the mean calculated differs from mu in one t test of class_ages

[@DataInsights2001]

Nice! Test control analysis is good for if promotion programs apply to test group and test whether there is any difference? Also nice to know that which direction to test, whether it is two tailed, left tailed, or right tailed? Also need to consider Type 1 error and type 2 errors?

[@neelroy3]

which statistical test can be used to find difference between two groups' percentage values?

[@prashu25925]

can we perform hypothesis testing on multiple column data?

[@itzmekallam7277]

how can 1.13 < 0.05, at 10:10 , is it mistake or just Krish Naik logic

[@NishatJillani]

At the end it shows -13 in power which you actaully missing . so how -13 power is greater then 0.05 even.

[@gunjanchaturvedi-m1b]

Let’s suppose 5 years ago, the average cost-per-person at a cafe was 300, has it changed now. (perform hypothesis testing to conclude). how to solve this which test need to perform here ..

[@kusumakatamneni3404]

Hi Krish we have to perform T-test in between two population is meaningful but how can we do on one population? How can we get significance difference on one population?

[@i_black_hawk]

if you need to test effect of drug on a sample of population then same sample of population needs to be taken before and after drug dose . This is done to reduce bias. And in such type of stats modelling we use paired sample t test

[@i_black_hawk]

Another example you can take is effect of workout on weight of a person. You need same sample of people whose weight were recorded before work out

[@kusumakatamneni3404]

@@i_black_hawk tanq so much

[@nishabhatt5268]

Thanks for sharing this video it is very helpful, can you please advise on how to integrate python script for p value in power bi? Many thanks

[@ankita684]

Thank you for this video Krish. In the two sample t test , 'ttest_ind' function you have taken equal _var to be 'False'. The code reads as : _,p_value=stats.ttest_ind(a=classA_height,b=ClassB_ages,equal_var=False). Shouldnt the 'equal_var' be True? As T test assumes that the populations have identical variances by default. Could you please check once. Thanks

[@munishasharma5265]

Hi Krish, I hope you are doing, Awesome explanation. However I have One Question What we will do after the analysis of these P test values?

[@piyushsaurav5791]

Hypothesis testing has been used by analysts to make inferences about the population .These tests are done to answer business questions eg . A/B testing ( version A is better than version B etc)

[@adityamahimkar6138]

I'm not an expert bt I think on a dataset we can make lot of changes statistically and train a model on the data bt using such test we can in the first look out of corrections in the data before training thus saving time and computation power. Do correct if I'm wrong, it just a null hypothesis 😅 :)

[@prasadrajmane4696]

Thank u sir

[@hvchetan1]

How do we get to know whether the test is one tail or two tail test... How python interpret this thing whether it's a one tail or two tail test?? As we are not specifying that thing .

[@sunilpatil1923]

Hello, thanks for the detailed explanations, Why p value is 5% only? Why can't it is 10% or 8% or any other value? Pls clarify.

[@krishnaik06]

It is decided before itself and yes it may changes...this value is decided by domain expertise

[@Andynath100]

Look up Statistical Hypothesis testing(Inferential Stats), It depends on the confidence level (alpha) for the test. In stats this measure of likelihood is taken as .05 (5%), .01(1%), .001(.1%) or in simpler terms what is the likelihood that the alternate is true if we assume the Null to be true. The t test calculates a p value and if it is less than alpha (.05 or 5% in this case) we reject the null because the probability of getting this sample if the null is true is very small (magnitude of p value). Or in simpler terms it's not possible (or very unlikely) to get this sample by chance. If you want a better explanation please follow the Udacity Intro to inferential stats course, its free.

[@abhinavjain5561]

Sir in 1sample t test second example,we take the mean of classA_ages as 30 but in next step it is coming 46 so what about that

[@shashikantrrathod3617]

Hello Krish, What are the limitation of linear statistical test? why we choose non-linear classifier over linear classifier?

[@rsinh3792]

Sir reviewer has asked me this question I don't know how to address it, can you please guide me "Use some statistical significant test such as T-test or ANOVA to prove you validate the proposed diagnostic model on patients and quality improvements of your method". I have two datasets. Dataset 1 was used to train the model and dataset 2 was used to validate the trained model. I have trained the ML model deployed it and Validated it on new data and presented the results. Actually, I have understood the question. Shall I apply the statistical test between the performance metrics of trained model results and validation results? Please help me, sir.

[@aksaini9063]

Sir if two independent feature are highly positive correlated or highly negative correlated. What is the best solution for this ?is the right to drp the one feature?

[@itplacementprep]

Summary : A One sample t-test tests the mean of a sample group against a known population mean. Two sample t test, An Independent Samples t-test compares the means for two groups. A Paired sample t-test compares means from the same group at different times (say, one year apart).

[@pramitbanerjee4381]

Why the classA_ages.mean() and classA_ages.var() not equal and what is the role of mu if it is not equal to classA_ages.mean()?

[@santiagorey1382]

very good video but please tell me, if the two features are statistically different, p_value> 0.05, then does that mean we should discard or keep that feature?

[@autismblessingindisguise]

Could not see the next video on chi Sq test implementation using python.. Please load it soon. Thanks

[@tannurohela6192]

Great explanation sir, but I didn't get why are we using np.random.seed() ? Can anyone please help with the seed thing.

[@meenadalvi9743]

@5.43 the p value is greater than 0.05 so in that case we should accept the NULL hypothesis.Correct me if am wrong.

[@Abhishek-st4mu]

same here, @10.00 i confusing on that statement, how can 0.05 is greater than 1.139

[@sohinibanerjee9617]

In the second one sample t test how is p value of 1.13 less than 0.05? Can someone please explain.

[@NishatJillani]

t the end it shows -13 in power which you actaully missing . so how -13 power is greater then 0.05 even.

[@zionramdinthara8403]

Hi Krish, can i compare two different population using t test. I want to compare the height of plants overtime with controlled and uncontrolled temperature. I actually have different datasets for both. Please help

[@dipk.mishra]

Sir , How do u decide whether Null is there is no difference ? Is there any logic behind?

[@ammar46]

Height and age doesn't follow Poisson distribution, it follows normal distribution.

[@hemantsharma7986]

Is one sample t test and one tail t test same?

[@mohinimarathe8769]

GOD OF STATS :)

[@dharamjeetsingh2936]

Krish i have a 3 years experience in business resilience analyst but we only use Excel not python SQL tableu. Do u think i have an advantage for becoming DS

[@sohinimitra5131]

In the first example of ages, you passed the expected NULL hypothesis value as 30 [ttest(ages_sample,30)]. Shouldn't it be 0? [ttest(ages_sample,0)]. Since The NULL hypothesis states the difference between mean of population and sample is 0. Why is the population mean passed there? Also, in many scenarios we will not have access to the population mean too.

[@lancelotdsouza4705]

pls explain this on a real dataset

[@c.dharmeshwaran3470]

In the 2 Sample T-Test, Were the samples/groups are selected from the same population or from 2 different populations?

[@soumyaranjansahu4262]

Hi Krish , In this School age problem, You have taken Sample Size=60 which is more that >30 .Hence shouldn't you calculate the P value on based on z- distribution rather than t-distribution?

[@amansinghrathore8308]

The t-test can be applied to any size (even n>30 also).

[@manitachakraborty2348]

can u please solve this T test problem without python

[@arpitjaiswal5972]

Why T test is used? Because there is no information given for Population SD so Z can't be used. If population SD was given then use Z test t distribution is Normal distribution / chi square. Check the formula you will be able to find the realtion

[@anupamjamatia]

hi, your tutorial is great, but I have a doubt regarding the statistical significance in this scenario -- if I do train data on Lang0 language and generate a model. afterward using the Lang0 model I do testing on other languages like Lang1, Lan2...Lang5 used different algorithms like AlgoA, AlgoB, and AlgoC and got the accuracy. so in that case is it possible to do the statistical significance test? no cross-validation is done while training. Say I have Lang Algo1 Algo2 Algo3 Algo4 Algo5 Lang1 80 32 95 93 96.67 Lang2 88 11 98 97 92.51 Lang3 49 12 76 80 72.75 Lang4 81 2 95 94 77.7 Lang5 81 43 95 96 94.95

[@adidbaker7607]

hey guys ive got a doubt in first one sample t test he said he is rejecting the null hyp when the p value is 0.740 which is higher than 0.05 ,so is isnt he supposed to accept the null hyp??

[@richasharma7968]

I have the same doubt. Can anyone explain it?

[@NishatJillani]

t the end it shows -13 in power which you actaully missing . so how -13 power is greater then 0.05 even.

[@pratikbambulkar8981]

But why we used hypothesis for ML?

[@sunitam1025]

sir, can you provide pdf of this

[@OnkarSingh-rg5jp]

Sir, in what case do we divide the p-value by 2?

[@siyabongamyeza5315]

He was supposed to divide the p-value by 2 since its a two sided test. Two sided test occur when you use the word "difference". When it is one sided, i.e either less than or greater than, you do not divide alpha by 2

[@lopamudrachandra2493]

Thank you so much for your video. Your channel is really helpful for students who cannot afford to online courses. I would like to know if I join your 59/month membership will it help me learn better on Data Science overall?

[@ankeshsingh2576]

If you execute the function ttest_1samp(), p_values keeps changing after every excution, varying from 0.05. How can we fix it ?

[@shreyasaxena5169]

If you execute random.choice then it will resample and change mean accordingly. For same sample , p value cannot vary.

[@questforprogramming]

Fix a number in random state.

[@utsavroy5346]

What if I reverse the assumptions? I means if H0 becomes H1 and vice versa. In that case how to move ahead?

[@akashprabhakar6353]

Yes you can do but u need to ensure that null hypothesis statement is chosen in such a way that you can conduct the experiment based on that null hypothesis. For example, you observation is that you got 10 head on 10 coin toss. Now you want to check if the coin is biased or not. Now, If you take Ho(null hypothesis ): coin is biased...then the problem is how will u find the p value or conduct the experiment ..bcz the coin can be biased with any probability And suppose you take Ho : coin is unbiased ...means probability of getting 10 heads on tossing the coin 10 times is : (0.5)^10......as probability of getting one heads is 0.5 for single toss when ""coin is unbiased"" Now u will get the values as 0.00097

[@devmani100]

Since you are dealing with the sample size and not the population, the relationship you might be getting from the sample may be due to random chances. The idea behind the null hypothesis is that relationship you are observing in the variables are due to randomness. S, my null hypothesis is always of the form that, "There is no relationship between the selected variables. This is what I have derived from all the sources from the StatsLand :D . Please correct me if I an wrong.

[@pratikbhansali4086]

What are we even achieving by doing one sample t Test

[@thousandsunny100]

ttest, p_value = ttest_1samp(covid, 30) TypeError: 'module' object is not callable

[@ppriyesh30]

Sir, this is unfair...I am just trying to build concepts of Data science and I come to know that there has been some term used which are totally new. Some times you import preprocessing, sometimes model_selection, sometime metrics, and, now, import maths..and poisons distribution and scipy stats.. please let us know when to choose what..Thanks..

[@ppriyesh30]

Specially please help us with the scikit learn library..remaining I guess has not that much importance

[@ManishKumar-qs1fm]

Plz corr explaine in details , bz m confuse in this

[@adeyinkaAdedejiNaMe]

Educative video but you calculated the population to be 30.4375 why did assume the population mean to be 30

[@shrikantdeshmukh7951]

Poisson distribution it's not poison distribution

[@BlueSkyGoldSun]

Fix the mistake ,how come 1.13 is less than 0.05?

[@karanbisht6359]

it was 1.12 *10^-somthing

[@SoumyaDasgupta]

Krish likes WWE. My Man

[@aws6143]

dimaag ho to essa ho jinda to pappu bhi h

[@vaibhavberiwal]

Watch khan academy videos for a more intuitive and in-depth understanding of the concepts :)

[@rsinh3792]

Sir reviewer has asked me this question I don't know how to address it, can you please guide me "Use some statistical significant test such as T-test or ANOVA to prove you validate the proposed diagnostic model on patients and quality improvements of you method" I have trained the ML model deployed it and Validated it on new data and presented the results. Actually, I have understood the question. Shall I apply the statistical test between the performance metrics of trained model results and validation results? Please help me, sir

[@001Debjeet]

1st

[@manikantasai721]

i too ...will do the same once upon a time .

[@KFarid-ri9fe]

Thhh,,,”

[@adityaprakashyadav4716]

Not upto the mark.

[@sasssystory]

Do you have video on f test anova?