I'm not sure what the interviewer was after but when facing a poorly preforming model these are the things I consider and an answer I would love hearing as an interviewer (been interviewing for a ...

This is one option: There is zero correlation for groups within vertical lines. But overall correlation is positive.

Yes. The second format fits a straight-forward classification problem. We have multiple classes(Red/Yellow/Green/Blue) and text based features. There are two main components needed in order to creat ...

No, similar distribution of Y's does not entail poor classification results. See this for example, two features X1, X2 and binary response variable(colored green and red). Clustering results are ...

Look into Gaussian mixture models. You basically assume an existence of 2 sub populations in your data. Fit a model with relevant parameters per each of the populations. Then using the parameters ...

I can think of a formal technique fitting this situation. semi-supervised learning. In the heart of this technique lays the assumption that all data(labeled and not labeled) have a common structure ...

This is a part of the feature engineering stage in modeling. Two key concepts in features engineering are: 1) Features should represent reality(or our assumption about reality). 2) Features different ...

PCA is a decomposition process. Meaning it takes an existing vector space and transforms it into another vector space. If i understand correctly, you chose the first N components of the transformed ...

One possible explanation might be a result of the game setting. Regardless of the cell the player is in - if the probability($P$) of it winning equals(or less) the probability of it meeting an enemy(...

The modeling strategy suggested in the paper refers to temporal representation(both frequency and context) of words. From what I understand, they attempt to learn the changes in these representations ...

I'm not sure if this is what you aiming for but this is one approach that might be suitable. Start with modeling the over 6 error as a probabilistic event. I.E, lets say that for each measure event $... View answer Accepted answer 1 votes I did the following calculation and got a different y value for x=0.25. Intercept and slope are similar to OP's question. x = (0.25, 1, 2, 2.75, 4.25, 6.5, 8, 13.25, 16.25, 19.25, 19.75, 26.5, 31, 37.... View answer Accepted answer 1 votes T-Test assumes samples are independent of each other. In the first case where each sample is identified by a user-date tuple, you might be violating this assumption. For example imagine the hypotheses ... View answer 1 votes As I understand it you are facing a problem of ordering(and then ranking) an unsupervised set of users. These type of problems can be tackled using a clustering algorithm. There are many such ... View answer Accepted answer 1 votes A clear cut answer would depend on the time series, actual transformations and clustering algorithm you have in mind. Reason is most clustering algorithms are based on different similarity function ... View answer 1 votes There is a concept named Asymptotic Normality. In our case means that over high enough number of estimations, the estimators distribution is normal. Technically this derives:$\hat{\beta} \sim \...

I know of three possible problem settings: 1. Detecting contextual anomalies in the time series - the anomalies are individual instances of the time series which are anomalous in a specific context, ...

Handling NaN values belongs to the feature engineering part of developing machine learning models. Different types of models make different assumption about the underline features distributions and ...

I would attempt modeling this type of revenue data as a time-series problem. Since you assume some sectors will go nil, you effectively assume the data is not stationary(average changes across time). ...

PCA can be used as a step for pre-processing the model features. From my experience: Pros: Reducing dimensionality. Reducing noise and possibly improving model performance. Cons: Sensitivity to ...

Its common to assume that the distribution-variance of the sample and that of the entire population are similar, so no surprise you didn't find an answer quickly. None the less you can use this ...

As for the number of common friend i suggest using the Jacard index. Its basically the ratio between shared friends both friends.

Some things to consider: Saying your sample size is constant, then each additional feature n will reduce you generalization by log(n) factor -- you can find a good explanation for this here: http://...

One potential direction is using contingency table. It is a good way of 'modeling' the relations between frequency distributions of more than one variable.

One way addressing problems where you have little(or no data) is 'replacing' it with priors. Bayesian modeling has the priors concept at its core. If your using python take a look at pymc(and start ...

My two cents: I would frame this as a classification problem where target label is 'died/survived'. Features are demographics, physical and transportation method. Assuming I have enough good data, ...

Using log transformations and lagged features are two separate techniques that could be combined together. In linear models, when regressors ${x_1, x_2..x_n}$ have multicollinearity one possible ...

As I understand it, your question is consisted of two smaller questions: What is the needed sample size to evaluate my classification model(algorithm)? How do I select my evaluation(labeled) data ...