data_science_in_julia_for_hackers
data_science_in_julia_for_hackers copied to clipboard
Chp 06 Review
Poor quality image
- [x] done
Give more details
- [x] done
Change
"To simplify, we approximate the escape velocity as:"
To something like:
"Assuming the planet is spherical and taking into account only the gravitational effect of Mars" (or whatever the assumptions are)
Otherwise is easy to get distracted thinking about what the approximation actually is, how crude it is, etc
format issue
- [x] done
"## Proyectile Motion"
Simplify
- [x] done
Change:
"...the x normally is parallel to the ground and the y axis is perpendicular,..."
To:
"...the x axis is parallel to the ground and the y axis is perpendicular,..."
Typo, ahve
- [x] done
So, in the example we ahve just explained,
Typo, and --> an
- [x] done
... suppose we only keep the measurements with and initial angle θ~45°
Use the same name in text and code
- [x] done
Suppose we did the experiment and we have measured then the 5 points, Δx and Δt, shown below:
Reword
- [x] done
This is very hard to read:
Using the equations of the trajectory, when the stone hits the ground, y(t) = 0, since we take the start of the y coordinate in the ground (negleting the initial height with respect to the maximum height), so finding the other then the initial point that fulfill this equation, we find that:
Typo
- [x] done
So, solving for v0m,
Change label or omit values and labels on y-axis
- [x] done
Change : "Probability"
Into: "Probability density"
Number equations
- [ ] done
It will be easier to follow
Out of the blue
- [ ] done
The sentence "So, the model we are going to propose is a linear regression. A linear equation has the form:"
is not well connected with the previous paragraphs. It is not clear why we need a linear model at all. The transition from mechanics to statistics should be smoother
Maybe you can invert the order of the story
- We want to scape from mars
- We need to find out the scape velocity
- For that we need to find g_mars
- We realize x = f(g, t), so we can throw stones to find g!
- But measurement are noisy, hence we create the model x ~ Normal(f(g, t), σ)
- We now try to find f from what we remember from high-school physics.
- We got it! we collect a few datapoints
- We explore a few priors.
Justify Priors
- [ ] done
say something like know it has to be positive and and less than g_earth, which is 9.8, and can round up to 10.
Do the same for the other two priors
Doubt HMC/ Turing good practices
- [x] done
Why are you using HMC sampler instead of NUTS or other similar adaptive dynamic HMC methods, the parameters you are passing are standard parameters? Standard for unidimensional posteriors?
I recommend to use NUTS instead of HMC even for simple models. I assume NUTS parameters are optional in turing, so you don't need to explain what epsilon or tau are (that you are not currently not explaining)
rewording
- [x] done
Change:
Now that we have a good understanding of the equations and the overall problem, we are going to add some difficulties and we will loosen a constrain we have imposed: Suppose that the device employed to measure the angle has an error of 15°, no matter the angle.
We want to know what are the most convenient angle to do the experiment and to measure or if it doesn’t matter.
to something like:
Let us remove a the angle constraint and add complexity to the problem. We will consider that the device measuring the angle has a 15 degree error, no matter the angle.
There is a most convenient angle to do the experiment? Or it doesn’t matter?
Discuss the posterior with angle uncertainty vs the one without it.
- [ ] done
Consider a new plot with them side by side or use overlapped. Discuss mean and standard deviation or HDI (or some other interval)
logarithmic --> LogNormal
- [x] done
first with a normal distribution and then with a logarithmic one.