felipewhitaker/decision_and_risk_analysis

Optimizing coverage and cost

Hi there, I'm Felipe!

I am very interested in leveraging data to solve challenges.

• 📂 This is my portfolio repository with a quick explation of public projects I have done!
• 💻 I mostly use Python and its amazing packages, which I used to write a simple package for transforming Time Series to a tabular format for Neural Networks: tabular-time-series
• 🏛️ I have used Julia for optimization problems during my Masters. Check out Machine Learning Under a Modern Optimization Lens and Decision and Risk Analysis repo.
• 🌱 I love learning new concepts and contexts through challenges
• 💬 Ping me for any data related challenge and I will love listening about it!
• ❤️ Fun fact: I love 🎶 music and 🚴 cycling!

🎓 Education

I'm an Operations Research Master student at the Industrial Engineering Department of PUC-Rio where also completed my BsC in Industrial Engineering with minor in Risk Analysis.

💼 Experience

Currently I am focused on my Master's dissertation, but previously I have worked with Innoavation, Research, Processes Automation, Data Science and applied Machine Learning: you can check all of it on my LinkedIn profile.

Feel free to connect with me and reach me!

decision_and_risk_analysis's Issues

Increase the sample size gradually to understand if installations and allocations converge

Something interesting to check is if the solution converges to some number of intallations and allocations, and a way to check for this is to vary the sample size.

The code below is the idea for it: read a sample, calculate the variables and then, considering a step and the maximum size, optimize for different sizes of samples.

sample_number = 0
# response distance
r = CSV.read("data/cooked/distances_$(sample_number).csv", DataFrame, skipto = 2, index = false)

# # population to be served at demand node i, using u type at t time
# d = trunc.(Int, rand(Uniform(5, 30), I[end], U[end], T[end]))
crimes = CSV.read("data/cooked/sample_$(sample_number).csv", DataFrame)
#  this is SUPER sparse (there are only I values in J * I * T matrix)

cov_cost_, vOpt_ = [], []
for sample_size in 10:50:1_000

    d = [((r.type_ + 1) == u && (r.shift + 1) == t) ? 1.0 : 0.0 for r in Tables.namedtupleiterator(crimes[1:sample_size, :]), u in U, t in T]

    # viable response distances
    N = [
        [j for (j, n) in enumerate(r[:, i] .<= Su[u]) if n] 
        for i in 1:sample_size, u in U
    ]

    perc_cov, total_cov, total_cost, xOpt, yOpt, zOpt = solve_(J, 1:sample_size, T, U, Pu, C, d, N, c_open, u_resource, 0.0)
    push!(cov_cost_, (total_cov, total_cost))
    push!(vOpt_, (xOpt, yOpt, zOpt))
end