I really like Python for data science uses so I was annoyed that SQL was an entirely different language that I would have to download more software for. Thankfully I found your series where I can do SQL from within Python cause apparently Python can do everything.
Here's a few hints from your profitable future. RN, GBDT & RNN expect bounded dataset as input; like words(all of dictionary, language context window), or chess(all possible moves, winning material). Given this, we understand that Price action = unbounded ($0 to $infinity, no fixed limits or sequence) and volatility = bounded(standard deviation). Any NN performs best in bounded datasets for training, validation & prediction. Therefore fundamentally Volatility is easier for CNN/RNN & HMM to predict. Jim Simons spoke of Hidden Markov Models being used to predict one of four regimes of volatility & momentum, then shifting option strategies that optimize performance in each regime. Straddle & Strangle is KEY - Options must be used for all operations as nature of prediction is expressed as an option. Best current approach is Dataset > (CNN+UMap+Timestep) + (RNN+Autoencoders) + RawData > HMM > Prediction. CNN+UMap performs hyper-dimensional feature optimization that is Timestepped, RNN+Auto passes temporal dependencies found in sequential dataset, then cumulatively is passed with raw features into HMM to parse hidden states to predict which t+1 market regime is expected. Given the predicted regime, one of four option spreads are taken against volatility for timestep interval. This can be dynamically adjusted to fit any timestep interval, 1d is assumed here. Good luck !
Just subscribed! I love the level you teach at in your videos. It’s slightly above the level of Statsquest but not too dense that I need to mentally prepare before watching. (No shade to Statsquest, two random events can be independently great).
one question, what if i have no idea of the correlation between the variants?, and actually, that's the thing that i want to find. Can i combine this method and also use the metropolis algorithm to find the values of mu and sigma and calculate p every iteration or something like that? thanks!
Allan Lichtman is by far the most accurate election forecaster. Correctly predicting 9 of 10 elections, the exception in 2000, although he insists Al Gore was the rightful winner. He’s likely to predict a Harris victory within a couple weeks. And he never pays attention to polls, analytics, or media punditry.
Beautiful Lady Kamala Harris to win! If the lunatic racist Donald Trump has been accused of crimes committed while he was president between 2016 and 2020, then why the Republicans have him again as their candidate for 2024 to commit more crimes? Only the far-right in America vote for Donald Trump who is not a politician and understands nothing about politics
Great vids!!! My question is to everyone including the creator: Is there any suggestion as to the order of videos I may want to watch them? Or these are just for completion of knowledge, where we watch just ANY topic that we need to know more of, like an encyclopedia?
Hey Ritvik, awesome video! I had a question. Transition matrix in HMM requires us to find the probability that the next word will have a state q given the state of the current word is k. But to figure out the probability we should have a corpus with POS tagging. But if we are using the HMM and viterbi algorithm to figure out the POS for a sentence then how can we have POS tagging already in place for a corpus? Isn't it kind of like the chicken and egg problem?
how can MCMC be used in the realm of stocks and finance? I’ve been looking into making a stockbot as a personal project and landed on MCMC as a viable option
Hi @ritvikmath, I recently came across your playlist on Time Series Analysis, and I found it incredibly insightful-thank you for that! I was wondering if you could make an additional video on ARIMAX, particularly focusing on how to incorporate exogenous predictor variables. That would be fantastic. Thanks again, and greetings from Germany!
Its great to be in times like these. Wonderful learning resources available on the internet for free. Some of my favourite learning resources: - 1) 3Blue1Brown 2) MIT Courses and the latest entrant: - 3) Ritvik Math Thanks for posting these videos!
Hi, I'm a pharmacist by training and at the moment doing a PhD in the field of pharmacometrics. I love your videos! It is just amazing that you are able to explain them to someone without a proper math or statistics background. In my field we are using quite a few data science concepts but in quite different (but at the same time similar) ways. I thought this video could be a great opportunity to share this with you (so you would keep explaining everything in such a simple way). In pharmacometrics, we are working with clinical trials data and developing nonlinear mixed-effects models to explain changes in concentrations of drugs in the human body over time (we have multiple drug concentration measurements over time, together with patient characteristics available). Our models are sets of ordinary differential equations and most of the parameters in these models are treated as random effects parameters (we can not pool all the samples from all the patients together, but need to account for samples belonging to the same individual, and that patients differ from one from another). Bias-variance tradeoff comes in place in the last stage of model development (after general trends were accounted for with the system of ODEs), when we are trying to find covariates on model parameters (some patient characteristics, such as age, or some blood measurement). In contrast to your field, we do not have multiple sample datasets available (splitting is not an option since the dataset size is rather small) and it usually takes years after we publish our models that somebody actually tries to validate them on external data. We opt for complex models that contain all possible covariates when we want to make inferences on covariate effects, or more precisely - when we want to prove that dose adjustment is not needed (when we want to estimate these negligible parameters), but we opt for a parsimonious model when we want to use it for simulations (other clinical trials, same special populations such as obese people etc.). Also when we are talking about bias in oversimplified models, I believe in the first place we mean that over parameter estimates (and not predictions) are biased, since if we do not include 2 covariates, but only 1, and the two are correlated, estimate of one will also account for not including the second one (estimated effects seems to be higher than it is). After watching your video I finally understand why we are saying that simpler models are more predictive. Thank you!
Great video! Wouldn’t it make sense to always choose uniform distribution as g(s). Ofcourse it could be uniform within a large interval for practical purposes. What would be the reasons to choose any other g(s)?
@@bege5918Just because your cross-eyed side votes for people based on their laughter, doesn't mean that everyone does. Neither candidates laughter or lack thereof will affect my vote.
@@kentreitzel5524 Even your blind-eyed side wanted to get rid of her as VP...and she was the first dropping out in 2020...soooo extremely fake and cringe...so I'd be a bit surprized if that fake enthusiasm is enough. We'll see...the importance of the passage of time will tell...😂
There is no need for polls or Prof Allan Lichtman's ridiculous 13 keys or astrology/psychic/data scientist predictions because population majority with below average IQ who don't give a damn if there is no food in their fridges, no gas in their cars and the illegal immigrants camping on their front yards, their sons and daughters in the army dying in unnecessary wars, their under-aged children having sex-change surgeries, but too excited to be part of the history making event of electing the first ever black female POTUS. It is super easy to predict a Harris landslide even though the whole country know that she is totally hopeless. She will win only because she is in the right place at the right time. Hadn't Biden stepped down just in time she would never have a chance.
How does this property of a vector (eigen vector) remains in the same dimension even after transformation (by A) helps in some problem solving (related to ML)?
Can you do a similar kind of analysis (with the polling datas) with the previous voting to see the trust level of the polling for that election and apply that to our current analysis?
The outcome aligns with my own analysis on the topic, funny enough it seems the election will come down to Pennsylvania which is 19 votes. So it is very likely your model is correct in this case.
If you place such little faith in the polling data, shouldn't you also put even less faith in the voting data because it is even more outdated? I feel assuming that increasing one means decreasing the other is somewhat flawed. I understand it from a mathematical view, total probability being 100%. But not having faith in polling data should be considered relative to faith in voting data. Thoughts?
I will definitely rewatch to help my own understanding, but I would love if you could cover the bayesian updating process in a stand alone video. That concept has always somewhat eluded me :/ Phenomenal work as always!
@@PR-cj8pd Because each Electoral Vote may not be equal like different states most of us understand the vote is counting the number and estimating what factor affects the vote it means if u work on Electoral Votes for output it can be skewed
Nice analysis man! There are two things we all know: 1, the elections are based on few swing states, or you can even say, swing counties. 2. some polls lies to manipulate public opinion or suppress some likely voters.