Lecture Improvement Project

[HumphreyYang]

When running through the list, please follow the following workflow:

1. Please assign yourself to one of the lectures below by adding an @{GitHub username} next to a lecture.

2. First, read each lecture carefully and check for typos and mathos (mathematical errors). If a derivation has a gap, consider adding an exercise to help fill it.

3. Rewrite the code in JAX where beneficial or feasible. One useful criterion is to check whether there are extensive loops or if the code is accelerated with numba.

4. Submit one pull request (PR) per lecture and tag @mmcky and @HumphreyYang for review.

Note

Sometimes there are good reasons to implement a single change across multiple lectures. If you think this applies
then please coordinate with @mmcky and @HumphreyYang, as these PRs can cause merge conflicts and complexity.

5. Note any questions or concerns in the PR. We’ll gather these into a bi-weekly summary/news feed for the group.

6. Synchronize updates to the Chinese version by tagging the Translation (English -> Mandarin) in Projects and marking the status as 'To do'.

Intermediate Series

Tools and Techniques

• sir_model
• linear_algebra
• qr_decomp
• eig_circulant
• svd_intro
• var_dmd (there is no simulation in this one)
• newton_method [jax] (There is a JAX version here). @kp992 - [newton_method] Update lecture code and fix typos lecture-python.myst#564

Elementary Statistics

• prob_matrix
• stats_examples
• lln_clt @mmcky
• prob_meaning
• multi_hyper [jax] (Uses jit + prange for combinatorics/MC) @HumphreyYang
• multivariate_normal
• hoist_failure (code needs to be cleaned up)
• back_prop [jax] (already in JAX -- it has some overlaps with @jstac's material on AI and economics and this lecture. We can try to integrate them into one)
• rand_resp
• util_rand_resp @HumphreyYang ([util_rand_resp] Update reference style across lecture and correct typos lecture-python.myst#621)

Bayes Law

Note for this section, there are discrepencies in terms of tools we used. Consider unifying them and write a introductory section for numpyro in bayes_nonconj?)

• bayes_nonconj [jax] (used both pyro with torch backend and numpyro with JAX backend) @kp992 - [bayes_nonconj] Update lecture lecture-python.myst#545
• ar1_bayes [jax] (used pymc and numpyro) ([ar1_bayes] Removed PyMC dependency and style-sheet compliance lecture-python.myst#569) @HumphreyYang
• ar1_turningpts [jax] (used pymc but not numpyro) ([ar1_turningpts] Removed PyMC and changed the style  lecture-python.myst#584) @xuanguang-li

Statistics and Information

(This chapter is currently being updated by Tom. Please refrain from reviewing it.)

• likelihood_ratio_process
• imp_sample [jax] (Uses jit/vectorize/prange for MC at the moment) @HumphreyYang ([imp_sample] Convert code to JAX and check stylesheet compliance lecture-python.myst#620)
• wald_friedman
• wald_friedman_2
• exchangeable [jax] (Numba vectorize/jit; easy JAX convertion.)
• likelihood_bayes
• mix_model [jax] (Currently mixes Numba and NumPyro; standardize on JAX.)
• navy_captain

Linear Programming

• opt_transport
• von_neumann_model

Introduction to Dynamics

• finite_markov
• inventory_dynamics [jax] (There is a JAX version here) - @kp992 - [inventory_dynamics] Update code to JAX and latest style guide lecture-python.myst#623, [inventory_dynamics] Update code to latest style guide and typo fixes lecture-jax#237
• linear_models
• samuelson @bishmaybarik
• kesten_processes [jax] (There is a JAX version here) @kp992 - [kesten_processes] Update Kesten processes lecture to latest code version lecture-jax#233, [kesten_processes] Kesten Processes lecture update lecture-python.myst#591
• wealth_dynamics [jax] (There is a JAX version here)
• kalman @xuanguang-li ([kalman] Update to JAX lecture-python.myst#611)
• kalman_2 @xuanguang-li ([kalman_2] Update to JAX lecture-python.myst#644)

Search

(Code in this chapter is similar in nature, consider convert the chapter to JAX)

• mccall_model [jax] (Noted in #505) @shlff
• mccall_model_with_separation [jax]
• mccall_fitted_vfi [jax] (Fitted VFI with jit/jitclass, we have a JAX version noted in #502)
• mccall_correlated [jax] @HumphreyYang ([mcall_correlated] Convert to JAX and content checks lecture-python.myst#616)
• career [jax] @HumphreyYang ([career] Convert to JAX and content checks lecture-python.myst#617)
• jv [jax] (numba jit in operator. Break up operator factory) @HumphreyYang ([jv] Convert to JAX and content checks lecture-python.myst#618)
• mccall_q [jax] (Very similar to the JS models above, should be a good target) @bishmaybarik
• odu [jax]

Consumption, Savings and Capital

• cass_koopmans_1
• cass_koopmans_2
• cass_fiscal
• cass_fiscal_2
• ak2 [jax] (AK model uses big loops that can be accelerated by JAX.)
• cake_eatingproblem @longye-tian
• cake_eating_numerical @longye-tian
• optgrowth @longye-tian
• optgrowth_fast [jax] (This lecture teaches Numba specifically, consider converting to JAX entirely) @HumphreyYang
• coleman_policy_iter [jax] (Time-iteration; clear @jit usage; good JAX target) @longye-tian
• egm_policy_iter [jax] (egm is implemented in JAX here) @longye-tian
• ifp [jax] (Uses numba at the moment) @longye-tian
• ifp_advanced [jax] (Uses numba at the moment) @longye-tian

LQ Control

• lqcontrol @matheusvillasb
• lagrangian_lqdp
• cross_product_trick
• perm_income
• perm_income_cons
• lq_inventories

Multiple Agent Models

• lake_model [jax] (Lake model class with @jit methods; break up using JAX) @HumphreyYang ([lake_model] Update lecture to JAX and check compliance to style sheet lecture-python.myst#589)
• rational_expectations
• re_with_feedback
• markov_perf
• uncertainty_traps
• aiyagari [jax] (There is a JAX version here) @HumphreyYang ([aiyagari] Update lecture with JAX and check for stylesheet compliance lecture-python.myst#590)

Asset Pricing and Finance

• markov_asset [jax] (There is a JAX version here) @xuanguang-li ([markov_asset] Update to JAX lecture-python.myst#654)
• ge_arrow @xuanguang-li
• harrison_kreps

Data and Empirics

• pandas_panel
• ols [jax] (There is a related JAX lecture here)
• mle [jax] (There is a JAX version here) @kp992 - [mle] Update lecture with latest code and style lecture-python.myst#572, [mle] Update JAX code to latest version lecture-jax#232

Auctions

• two_auctions ([two_auction] Lecture review with math and warning fixes lecture-python.myst#556) @HumphreyYang
• house_auction

Advanced Series

Tools and Techniques

• orth_proj
• stationary_densities
• muth_kalman
• discrete_dp

LQ Control

• cons_news
• smoothing
• smoothing_tax
• markov_jump_lq
• tax_smoothing_1
• tax_smoothing_2
• tax_smoothing_3
• lqramsey

Multiple Agent Models

• arellano [jax] (There is a JAX version here)
• matsuyama
• coase
• match_transport

Dynamic Linear Economies

• hs_recursive_models
• growth_in_dles
• lucas_asset_pricing_dles
• irfs_in_hall_model
• permanent_income_dles
• rosen_schooling_model
• cattle_cycles
• hs_invertibility_example

Risk, Model Uncertainty, and Robustness

• five_preferences [jax] (Analytic kernels with @njit)
• entropy
• robustness
• rob_markov_perf

Time Series Models

• arma
• estspec
• additive_functionals
• lu_tricks
• classical_filtering
• knowing_forecasts_of_others

Asset Pricing and Finance

• lucas_model [jax] (There is a JAX version here) @kp992
• asset_pricing_lph @Simon-Mishricky
• black_litterman
• BCG_complete_mkts
• BCG_incomplete_mkts

Dynamic Programming Squared

• un_insure [jax] (Not accelerated, but with big iteration of C)
• dyn_stack
• calvo_machine_learn
• calvo
• calvo_abreu
• opt_tax_recur
• amss [jax] (Heavy DP with jit/jitclass; good JAX target)
• amss2
• amss3
• chang_ramsey
• chang_credible

[jstac]

Thanks @HumphreyYang ! Great summary.

I learned a new word -- mathos :-)

All help is appreciated. If a lecture is not computationally intensive and it's written in numpy then we may as well keep it that way. As @HumphreyYang says, any lecture that's computationally intensive or accelerated by numba need to be considered as candidates for shifting to JAX. And all lectures will benefit from review.

[HumphreyYang]

To everyone reading this issue:

Lectures with the [jax] tag at the end are generally good candidates for JAX conversion in the first round. These are typically lectures that have been partially implemented in JAX, are similar to lectures already in JAX, or would benefit from being converted to JAX.

[mmcky]

@QuantEcon/lecture-review-modernization just a reminder to attach your @{github-username} against a lecture of interest.

For RAs that have recently joined the team, we are also starting to use the good first issue tag if you prefer to take a look at that list and take ownership of one of these issue. You can also tag yourself against a lecture and specify styleguide review to indicate you would like to read a lecture and apply the QuantEcon Style Guide for example.

[mmcky]

@QuantEcon/lecture-review-modernization -- once you are ready to open a DRAFT PR -- please copy+paste a link to the PR (after your name) in the table above. This makes reviews pretty convenient as we go down the list, the status of the link automatically updates once the PR is merged, and it links your PR to this issue. Thanks.

We will run through tomorrow how to open a DRAFT PR, and a few other GitHub tips.