| A | B | C | D | E | F | G | H | I | J | K | L | M | N | O | P | Q | R | S | T | U | V | W | X | Y | Z | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | Question Report | |||||||||||||||||||||||||
2 | Report Generated: | Jun 14, 2020 1:03 AM | ||||||||||||||||||||||||
3 | Topic | Webinar ID | # Question | Actual Start Time | Actual Duration (minutes) | |||||||||||||||||||||
4 | Artificial Intelligence and Deep learning Course by IIT Roorkee | 962 3602 6138 | 175 | Jun 13, 2020 7:33 PM | 217 | |||||||||||||||||||||
5 | Question Details | |||||||||||||||||||||||||
6 | # | Question | Answer(s) | Asker Name | Asker Email | |||||||||||||||||||||
7 | 1 | are you going to cover multi task network or multi task loss ? | I did not get your question ..could you please elaborate ..we will sover RMSE as cost function in this session | Arpit vijaywargiya | arpitvw16@gmail.com | |||||||||||||||||||||
8 | 2 | Why do we need Box plot when we have histogram for same visualisation? | live answered | Sugandhita | sugandhitap@gmail.com | Box plot is another way of looking at data especially on an interval scale | ||||||||||||||||||||
9 | 3 | in object detection we have multi task network, which detect obect and give bounding box, | Arpit vijaywargiya | arpitvw16@gmail.com | ||||||||||||||||||||||
10 | 4 | so there will be 2 losses one is for classification and other is for bounding box | Arpit vijaywargiya | arpitvw16@gmail.com | ||||||||||||||||||||||
11 | 5 | What information do we get out from skewness and modality? | machine learning algorithms can not learn properly if the data is tail heavy or skwed…we will discuss the reason in Training models chapter | Satyabrat Sabat | satya.jin@gmail.com | |||||||||||||||||||||
12 | 6 | Q2) generally we use Z-value (depends on median ) for P-value, can we drive same thing with mode , as it is robust for outliers | Arpit vijaywargiya | arpitvw16@gmail.com | ||||||||||||||||||||||
13 | 7 | what is the impact of non distributed feature on model creation? | live answered | sudhir shetty | sudhir.m.shetty@gmail.com | |||||||||||||||||||||
14 | 8 | *(depends on mean) | Arpit vijaywargiya | arpitvw16@gmail.com | ||||||||||||||||||||||
15 | 9 | *drive things from Median | Arpit vijaywargiya | arpitvw16@gmail.com | ||||||||||||||||||||||
16 | 10 | Hi, May I know which books we are referring please for current diagrams? This si something I would like to read in details too. I purchased O'reilly one on Hands-On Ml and great help connecting last two classed but these histograms etc not there. | These are the general concepts …we did not follow any book for the same. | Rajiv | krajiv.2018@gmail.com | |||||||||||||||||||||
17 | 11 | How the Variance is diff from SD | Standard deviation and variance both show spread. Standard deviation is the square root of variance | Rajeev | rajeev213149@gmail.com | |||||||||||||||||||||
18 | 12 | error miimise | Chinmay Athavale | chinmayat@gmail.com | ||||||||||||||||||||||
19 | 13 | same unit as data | sudhir shetty | sudhir.m.shetty@gmail.com | ||||||||||||||||||||||
20 | 14 | otehr wise data will get canceled | Dr. santoshkumar | ksantosh.11@gmail.com | ||||||||||||||||||||||
21 | 15 | beacause there is both + and - diff | 409992 | supriya.bms@gmail.com | ||||||||||||||||||||||
22 | 16 | because the difference can be negative and positive | Shalini Gupta | gupt.shalu1993@gmail.com | ||||||||||||||||||||||
23 | 17 | so that difference won't cancel | kunal upadhyay | kupadhy@gmail.com | ||||||||||||||||||||||
24 | 18 | ok, thank you | Rajiv | krajiv.2018@gmail.com | ||||||||||||||||||||||
25 | 19 | What is standard deviation? What is the difference between standard deviation and variance? | Standard deviation and variance both show spread. Standard deviation is the square root of variance | Anantpadmanabh Divanji | apgd14@gmail.com | |||||||||||||||||||||
26 | 20 | Squaring adds more weight to the larger differencs | Anmol Khopade | anmolck@gmail.com | ||||||||||||||||||||||
27 | 21 | This is important when points further than the mean are important | Anmol Khopade | anmolck@gmail.com | ||||||||||||||||||||||
28 | 22 | what is the diffrence between bias and variance ? | The bias error is an error from erroneous assumptions in the learning algorithm. High bias can cause an algorithm to miss the relevant relations between features and target outputs (underfitting). Whereas, the variance is an error from sensitivity to small fluctuations in the training set. High variance can cause an algorithm to model the random noise in the training data, rather than the intended outputs (overfitting). | Vikas Bhartiya | ghivikas@gmail.com | |||||||||||||||||||||
29 | 23 | Is mean and median being equal or closer to each other is the way of determining a distribution is normal or not? | Normal distributions are symmetric, unimodal, and asymptotic, and the mean, median, and mode are all equal. | Puneet Rsstogi | puneetrstg@gmail.com | |||||||||||||||||||||
30 | 24 | Why Normal Distribution is imp? | A lot of physical phenomenon follow normal distribution. We have built understanding and modelling capabilities for them. | Rajeev | rajeev213149@gmail.com | |||||||||||||||||||||
31 | 25 | does variation of mode vs median tell us anything | A difference in mode and median typically means there is some skewing in the data. It calls for checking for outliers, long or fat tails, etc | Anmol Khopade | anmolck@gmail.com | |||||||||||||||||||||
32 | 26 | how do we know if a dataset follows a normal distribution? | For normal distribution the mean, median, mode are equal. To test for noramlity we can find the skewedness of the data. | Nini Nursiah | nursiah.neelesh28@gmail.com | |||||||||||||||||||||
33 | 27 | what is z-value? | A z-score gives you an idea of how far from the mean a data point is. Tt's a measure of how many standard deviations below or above the population mean a raw score is. | VED | parmarvedpro5@gmail.com | |||||||||||||||||||||
34 | 28 | Why do we take root of it? | By taking the root the unit of error is same as the observation | Senthilnathan Ramaswami | senthilnathan.ramaswami@servicenow.com | |||||||||||||||||||||
35 | 29 | When do we use RMSE and MSE? | Hi Srini, this is a good question ..can you please post it on forum and we would explain it in detail | Srini Boddu | siliconfish@yahoo.com | |||||||||||||||||||||
36 | 30 | what is the relation between variance and z-value | z = (x - mean)/ (standard deviation) | VED | parmarvedpro5@gmail.com | |||||||||||||||||||||
37 | 31 | how many lines will the algorithm try? As there are infinite way to assign w)0 and w1? | We will learn this in upcoming chapters while discussing gradient descent | Nini Nursiah | nursiah.neelesh28@gmail.com | |||||||||||||||||||||
38 | 32 | What is W0 and W1 here ? | These are coefficients | Nitin Nigam | nknigam@gmail.com | |||||||||||||||||||||
39 | 33 | For different lines drawing, are we using different data? | No, the algorithm tries to find the line which best fits the available data. The best fitting line is the one with lowest RMSE. | Veeru(VeeraNancharaiah Javvaji) | jsriveeru@gmail.com | |||||||||||||||||||||
40 | 34 | I meant in which scenario, do you choose MSE and in which one you use RMSE? | I think it is already answered | Srini Boddu | siliconfish@yahoo.com | |||||||||||||||||||||
41 | 35 | i didn’t get why pam score is better ? | Anonymous Attendee | |||||||||||||||||||||||
42 | 36 | what does negative sign indicate in output of describe method ? | jia sharma | jiavidhi.sharma@gmail.com | ||||||||||||||||||||||
43 | 37 | what are the x axis and y axis in hist plot | X is the value and Y axis has the frequency of the value | Rajeev | rajeev213149@gmail.com | |||||||||||||||||||||
44 | 38 | More towards left - near 34 | Rajiv | krajiv.2018@gmail.com | ||||||||||||||||||||||
45 | 39 | what does bins represent? | Venkata Pradyumna | ivpradyumna@gmail.com | ||||||||||||||||||||||
46 | 40 | Latitude bimodal and longitude bimodal—> LA and SF? | Yes, more dense population in certain areas. | Domenico Fioravanti | nicodom@gmail.com | |||||||||||||||||||||
47 | 41 | riht skewed | Dr. santoshkumar | ksantosh.11@gmail.com | ||||||||||||||||||||||
48 | 42 | one outlier and skewed | Sourav Ghosh | souravghosh@hotmail.com | ||||||||||||||||||||||
49 | 43 | How the peak will help in observation? is there anything do you see in peak value? | The peak, skew etc are methods for us to understand the data better. Having an intuitive sense of what the data is and what effect will be goes long way in building a good model. | Rajiv | krajiv.2018@gmail.com | |||||||||||||||||||||
50 | 44 | LA = 34.0522° N, 118.2437° W SF =37.7749° N, 122.4194° W (yes I confirm) | Domenico Fioravanti | nicodom@gmail.com | ||||||||||||||||||||||
51 | 45 | What does capping at 15 means? | That the income was capped at 15 ..all the values greater than 15 was capped to 15 …Hope it answers your question | Aswin Sabaaree | aswin.sabaaree@innovatia.net | |||||||||||||||||||||
52 | 46 | what is the reason for spike? | Sugandhita | sugandhitap@gmail.com | ||||||||||||||||||||||
53 | 47 | does X has to be normally distributed or Y or both? | TYpically Y is. X usually is not | Sourav Ghosh | souravghosh@hotmail.com | |||||||||||||||||||||
54 | 48 | Caped means that the value grated then the caped value are converted to the caped value like median_house_value > 500000 will be converted to 500000 | DIvya Pathak | dev.feb88@gmail.com | ||||||||||||||||||||||
55 | 49 | why is it easier to perform ML on normal data? | We will learn this in the upcoming chapters | Nini Nursiah | nursiah.neelesh28@gmail.com | |||||||||||||||||||||
56 | 50 | How the ratio 80/20 was chosen? Is there a reason? | depends on the size of data ..we use either 80:20 or 70:30 | Domenico Fioravanti | nicodom@gmail.com | |||||||||||||||||||||
57 | 51 | why is outliers removed if they a good in numbers. Is it not a problem of data collection? | Outliers is often removed to make modelling easier. Outliers are a real thing in most data collection exercises | Krishna Mohan | kmiitan96@gmail.com | |||||||||||||||||||||
58 | 52 | How do we convert attributes to bell curve shaped ? | All attributes may not be convertibile to a bell shaped curve | Srihari M | srihariblr12@gmail.com | |||||||||||||||||||||
59 | 53 | How to decide %s taken fro Training set Vs Test Set? Like you took 80% for training and 20% for test set. what are various things I need to think for it. | Typically we choose 20% for the test set. For large dataset we can choose less % for test data. | Rajiv | krajiv.2018@gmail.com | |||||||||||||||||||||
60 | 54 | There is a terminology Out of Time validation. Is it same thing as Test set? | The out-of-time validation sample contains data from an entirely different time period or customer campaign than what was used for model development. Validating model performance on a different time period is beneficial to further evaluate the model's robustness. | Puneet Rsstogi | puneetrstg@gmail.com | |||||||||||||||||||||
61 | 55 | What does capping at 15 means? | Aswin Sabaaree | aswin.sabaaree@innovatia.net | ||||||||||||||||||||||
62 | 56 | how do you know which data goes to training set and which data will go to test set? | live answered | Stuti Rastogi | e0498211@u.nus.edu | We want the same type of data in training and test. The best way is often to do some random distribution | ||||||||||||||||||||
63 | 57 | Does the Caping means that the value grated then the caped value are converted to the caped value like median_house_value > 500000 will be converted to 500000 ? | That is correct | DIvya Pathak | dev.feb88@gmail.com | |||||||||||||||||||||
64 | 58 | @TAs - When do we use MSE and when do we use RMSE? | Srini, I am not sure if there is a perfect answer. Both of them are a measure of the error. The MSE may penalise the error more severely than the RMSE, which may help some times and maybe a detriment other times. | Srini Boddu | siliconfish@yahoo.com | |||||||||||||||||||||
65 | 59 | Why is training not performed on 100% of data? | Because you need to test quality of the model you create. If you used it for training, the model will usually show a good result on that | Nikhil Sharma | nikhilthemacho@gmail.com | If you train on 100% data, you will have no data left to evalute your model on how it is performing. | ||||||||||||||||||||
66 | 60 | As a human we can be bias but how come a ML algorithm is bias ? | If ML algorithm will learn from the biased data then ofcourse the final model will be biased :) | Vikas Bhartiya | ghivikas@gmail.com | |||||||||||||||||||||
67 | 61 | What are different mechanism to split data? is Hash most famous? | We will cover diff techniques now | Rajiv | krajiv.2018@gmail.com | |||||||||||||||||||||
68 | 62 | time based observations | Sourav Ghosh | souravghosh@hotmail.com | ||||||||||||||||||||||
69 | 63 | what is biased? | VED | parmarvedpro5@gmail.com | ||||||||||||||||||||||
70 | 64 | The training and test data might have same row if we use the ramdom permutation? | No the training data and test data are disjoint. | Nini Nursiah | nursiah.neelesh28@gmail.com | |||||||||||||||||||||
71 | 65 | but can't the splitting be done initially and then models be run with same trainign and test set? | Preedesh M | Preedesh@gmail.com | ||||||||||||||||||||||
72 | 66 | Why 42 is passed to random.seed? | It is just a random number, you can choose a different one. | Anantpadmanabh Divanji | apgd14@gmail.com | You can use another number. The results may vary a little. | We can pass anny number as randomm seed …but make sure to use that number through out your model training process | |||||||||||||||||||
73 | 67 | Seed can be any values or based on any observation we need to consider ? | It can be any value | Satya Sunil | kvv.satyasunil@gmail.com | |||||||||||||||||||||
74 | 68 | ok thanks Praveen | Anmol Khopade | anmolck@gmail.com | ||||||||||||||||||||||
75 | 69 | what exactly is Hash? | Puneet Rsstogi | puneetrstg@gmail.com | A hash is a function that converts one value to another. Hashing data is a common practice in computer science and is used for several different purposes. Examples include cryptography, compression, checksum generation, and data indexing | |||||||||||||||||||||
76 | 70 | what is the relative advantages of using md5 vis-a-vis the other one pls? | Typically any hash that randomises sufficiently will do the work | Sourav Ghosh | souravghosh@hotmail.com | |||||||||||||||||||||
77 | 71 | what is hash. Please elaborate | https://en.wikipedia.org/wiki/Hash_function Hashes are used to create a random representation of the data. This removes any systematic grouping like all data from the same area may be together. | Sugandhita | sugandhitap@gmail.com | |||||||||||||||||||||
78 | 72 | why can't we just select the last 20% of the dataset for test set? | The last 20% may come from the same area and may not be representative of the entire data set | Nini Nursiah | nursiah.neelesh28@gmail.com | Then the data used for training the model will not be inclusive. | ||||||||||||||||||||
79 | 73 | why do we need to append data at the end? | Aakash Sinha | post2aakash@gmail.com | ||||||||||||||||||||||
80 | 74 | can more comments be added into the python file to explain the code section ? reason is there are code sections which are alternative ways to achieve similar goal. | noted the feedback | Prakhar Prasad | prakhar.prasad@gmail.com | |||||||||||||||||||||
81 | 75 | do we need Id column to create if we use train_test_split() ? | If we are using scikit-learn function then no ..it will split based on the row index I believe | Nitin Nigam | nknigam@gmail.com | |||||||||||||||||||||
82 | 76 | what is the best way for getting the train and test data . Creating own function or using Sckit Learn Function | using scikit-learn function | DIvya Pathak | dev.feb88@gmail.com | You can do both. Use the Scikit Function is easier and usually better | ||||||||||||||||||||
83 | 77 | Can we see an example of sampling bias? | Hi Sourav, is your question answered now? Prof just explained it again for Jia’s question | Sourav Ghosh | souravghosh@hotmail.com | |||||||||||||||||||||
84 | 78 | sampling bias arises due to the way we collect the data right? | yes | Nini Nursiah | nursiah.neelesh28@gmail.com | |||||||||||||||||||||
85 | 79 | Like in CNN's cant we take entire data and split it into 80-20 ratio where 80% is training and 20% validation and we refine based on testing accuracy. Random just seems too uncontrollable even with larger dataset | Anmol Khopade | anmolck@gmail.com | ||||||||||||||||||||||
86 | 80 | how do you take a sample of 100 out of 200 million adult population? But I do realize that they take the samples, from the whole population set? What is the process to in corporate the whole population? | Its not possible to incorporate the entire population usually. So you try to sample a smaller set for your excercise. | Srini Boddu | siliconfish@yahoo.com | |||||||||||||||||||||
87 | 81 | This sampling bias exists when we accquired the data or it also exits while splitting test and train data? | We are going to show this | Puneet Rsstogi | puneetrstg@gmail.com | |||||||||||||||||||||
88 | 82 | how is stratified samples different from cluster samples | Krishna Mohan | kmiitan96@gmail.com | ||||||||||||||||||||||
89 | 83 | what is sampling bias in the example shown ? please explain | jia sharma | jiavidhi.sharma@gmail.com | ||||||||||||||||||||||
90 | 84 | I got my answer - Thanks | Srini Boddu | siliconfish@yahoo.com | ||||||||||||||||||||||
91 | 85 | How do we measure if a status is large enough? | Domenico Fioravanti | nicodom@gmail.com | ||||||||||||||||||||||
92 | 86 | If know data in advance, we can categories the income | Veeru(VeeraNancharaiah Javvaji) | jsriveeru@gmail.com | ||||||||||||||||||||||
93 | 87 | for future data, how can ? | For near future term we can assume that the income will stay in similar range. | Veeru(VeeraNancharaiah Javvaji) | jsriveeru@gmail.com | |||||||||||||||||||||
94 | 88 | *stratus | Domenico Fioravanti | nicodom@gmail.com | ||||||||||||||||||||||
95 | 89 | How do the prof determine the the income as the measure for stratified sampling? or did he pick up randomly? | Srini Boddu | siliconfish@yahoo.com | ||||||||||||||||||||||
96 | 90 | Wouldn’t it be better to use bins=[0., 2, 3.0, 4.5, 6., np.inf], to increase the size of the first stratus? | Domenico Fioravanti | nicodom@gmail.com | ||||||||||||||||||||||
97 | 91 | 2 to 6 has most | AG | abhijeetgadgil@gmail.com | ||||||||||||||||||||||
98 | 92 | so how to represent 2 to 6 in right manner | Hi AG, did not get your question ..can you please elaborate | AG | abhijeetgadgil@gmail.com | |||||||||||||||||||||
99 | 93 | If I look at bins we say 0 to 1.5 and then 1.5 to 3, but why are bins picked in 1.5 ranges in this example | There are a couple of way we can determine the optimum bin size. First, we find the smallest and largest data point, lower the minimum a little and raise the maximum a little, decide how many bins you need, divide your range (the numbers in your data set) by the bin size, and finally create the bin boundaries. Second is you can use Sturge’s Rule, which is K = 1 + 3. 322 logN, where K is the number of class intervals, N is the number of observations. | AG | abhijeetgadgil@gmail.com | |||||||||||||||||||||
100 | 94 | why was numerical converted to categorical? I didn't understand that part | Categorical was converted numerical, to enable modelling | Nini Nursiah | nursiah.neelesh28@gmail.com |