yeah, it makes up stuff and states it as fact, its entertainment value that I coaxed into doing something useful . when you question it, it usually corrects itself, but sometimes gets stuck in a loop repeating something thats obvious nonsense or invalid. for sheer novelty and hack value its interesting.
approaches to calibrating a Heston model
- Thread starter stochastix
- Start date
Keep in mind this is still in alpha or beta., the real innovation here is the semantic accuracy of the language model, not its ability or lack thereof to crawl the web. In many ways we are at the very beginning of technological innovation. The things I could dream up will be reality in a few years or a decade, I could fill books.
it just made a cool animated java program of a stick figure randomly waiving its hands hopping around the screen
I believe this is because it just gives the answer with the highest probability as if it is correct and we don't get to see the alternative possibilities. It is giving the answer as 100% even if internally the answer only has a 40% probability of being correct as there is just much uncertainty in the response. I wish we were able to see more of this under the hood but I can see how people wouldn't relate to the answers the same way.
The other weakness seems to be that it has to treat int/floats as strings and has no concept of time. It is understandably bad with the ordering of events. A good example is to ask it what advanced civilizations were around 10,000 years ago. It told me ancient Egypt was an advanced civilization around 10,000 years ago but that does make sense as the highest probability answer when there is really no good answer.
That regression problem it has to treat everything in the document as a string and then it is also keeping state in the chat window. I think if you tell it read that document the information at the end of the document will muddle the probability of the answers from the start of the document. It gave me this proof through dialog with it but I have no idea if it is correct, totally over my head.
"Sure! To prove that H is a projection matrix, we need to show that it has the following properties:
Symmetric: HT = H
Idempotent: H × H = H
Proof:
Symmetric:
Let's start by showing that H is symmetric. To do this, we need to show that HT = H. We have:
HT = (X(XTX)-1XT)T
= (XT)T((XTX)-1)T XT [using the transpose rule (AB)T = BT AT]
= X(XTX)-1XT [since (XTX)-1 is symmetric]
Therefore, we have HT = X(XTX)-1XT = H, which proves that H is symmetric.
Idempotent:
Next, let's show that H is idempotent, i.e., H × H = H. We have:
H × H = X(XTX)-1XT × X(XTX)-1XT
= X(XTX)-1(XT × X)(XTX)-1XT [using the associative property of matrix multiplication]
= X(XTX)-1XT [since (XTX)-1(XTX) = In, the identity matrix]
Therefore, we have H × H = X(XTX)-1XT = H, which proves that H is idempotent.
Since H satisfies both properties of being symmetric and idempotent, it is indeed a projection matrix."
It can show the proof in numpy too
import numpy as np
# Generate some random data
X = np.random.rand(10, 3)
y = np.random.rand(10)
# Calculate the OLS estimator of the coefficients
beta_hat = np.linalg.inv(X.T @ X) @ X.T @ y
# Calculate the hat matrix
H = X @ np.linalg.inv(X.T @ X) @ X.T
# Calculate the residuals
residuals = (np.eye(len(y)) - H) @ y
# Check if H is symmetric
print(np.allclose(H, H.T))
# Check if H is idempotent
print(np.allclose(H @ H, H))
It does print true true but I wish it would give a confidence score.
The other weakness seems to be that it has to treat int/floats as strings and has no concept of time. It is understandably bad with the ordering of events. A good example is to ask it what advanced civilizations were around 10,000 years ago. It told me ancient Egypt was an advanced civilization around 10,000 years ago but that does make sense as the highest probability answer when there is really no good answer.
That regression problem it has to treat everything in the document as a string and then it is also keeping state in the chat window. I think if you tell it read that document the information at the end of the document will muddle the probability of the answers from the start of the document. It gave me this proof through dialog with it but I have no idea if it is correct, totally over my head.
"Sure! To prove that H is a projection matrix, we need to show that it has the following properties:
Symmetric: HT = H
Idempotent: H × H = H
Proof:
Symmetric:
Let's start by showing that H is symmetric. To do this, we need to show that HT = H. We have:
HT = (X(XTX)-1XT)T
= (XT)T((XTX)-1)T XT [using the transpose rule (AB)T = BT AT]
= X(XTX)-1XT [since (XTX)-1 is symmetric]
Therefore, we have HT = X(XTX)-1XT = H, which proves that H is symmetric.
Idempotent:
Next, let's show that H is idempotent, i.e., H × H = H. We have:
H × H = X(XTX)-1XT × X(XTX)-1XT
= X(XTX)-1(XT × X)(XTX)-1XT [using the associative property of matrix multiplication]
= X(XTX)-1XT [since (XTX)-1(XTX) = In, the identity matrix]
Therefore, we have H × H = X(XTX)-1XT = H, which proves that H is idempotent.
Since H satisfies both properties of being symmetric and idempotent, it is indeed a projection matrix."
It can show the proof in numpy too
import numpy as np
# Generate some random data
X = np.random.rand(10, 3)
y = np.random.rand(10)
# Calculate the OLS estimator of the coefficients
beta_hat = np.linalg.inv(X.T @ X) @ X.T @ y
# Calculate the hat matrix
H = X @ np.linalg.inv(X.T @ X) @ X.T
# Calculate the residuals
residuals = (np.eye(len(y)) - H) @ y
# Check if H is symmetric
print(np.allclose(H, H.T))
# Check if H is idempotent
print(np.allclose(H @ H, H))
It does print true true but I wish it would give a confidence score.
Last edited:
I managed to get it to find the eigenfunctions of the laplacian of the lemniscate metric... something that isnt a standardly calculated thing.
https://github.com/crowlogic/arb4j/wiki/LemniscateMetricEigenfunction
https://github.com/crowlogic/arb4j/wiki/LemniscateMetricEigenfunction
You sound like a second year TA. You build a better model that earns you three beeps when microstructure costs you five as a price taker.
If your goal is to make money in vol -> arrive at an index volcorr and RR-model -> trade skew modality on index -> trade D1/D2>1 switch (term-structure) edge if vol-surface is flat. There are more index and SN ideas than I can trade. Or simply trade spec in vol.
If your goal is to make money in vol -> arrive at an index volcorr and RR-model -> trade skew modality on index -> trade D1/D2>1 switch (term-structure) edge if vol-surface is flat. There are more index and SN ideas than I can trade. Or simply trade spec in vol.
I appreciate it.. that isnt me, its just some stuff that was transmitted on the computer... it writes in the style of whatever it read.. was just messing around with its capabilities
those are good ideas, I've finally got my arbitrary precision library ready to use, I had problems with double-precision arithmetic being a limited factor before but it is not any longer,
so now I'm dusting off this Fourier inversion code since I know lots of fourier inversion methods exist but none have been so far written for use with arblib or arb4j
So you're trying an experiment. Cast pearls before ET swine and see if there is any uptake. My guess is there will be virtually none.
FFS you were just handed the keys and you are still worrying your github code? Put the fourier inverse code on the back burner, trade what you were just told, and come back to it when you can afford a Bently and an impluse vacay to Hawaii.stochastix said:
LMAO
I just got back from an impulse vacay to Hawaii, spent about 3 and a half months there
casting perls is the wrong metaphor to invoke btw
phrase [VERB inflects] If you say that someone is casting pearls before swine, you mean that they are wasting their time by offering something that is helpful or valuable to someone who does not appreciate or understand it.
even though the jargon is a bit dense i get the gist
phrase [VERB inflects] If you say that someone is casting pearls before swine, you mean that they are wasting their time by offering something that is helpful or valuable to someone who does not appreciate or understand it.
even though the jargon is a bit dense i get the gist