search

google / lightweight_mmm

LightweightMMM 🦇 is a lightweight Bayesian Marketing Mix Modeling (MMM) library that allows users to easily train MMMs and obtain channel attribution information.
https://lightweight-mmm.readthedocs.io/en/latest/index.html
Apache License 2.0
831 stars 172 forks source link

ConcretizationTypeError: Abstract tracer value encountered where concrete value is expected #198

Open alberto-molinaro opened 1 year ago

alberto-molinaro commented 1 year ago

Hello everyone,   I am trying to find the optimal budget allocation for two given budgets (for example 2000 and 3000).

Here is my code: 

########### MODEL INPUT PATH ############
#######################################
model_path = 'path_to_model.pkl'
media_scaler_path = 'path_to_media_scaler.pkl'
target_scaler_path = 'path_to_target_scaler.pkl'
 
 
########### LOAD THE MODEL ##############
########################################
mmm_model, media_scaler, target_scaler, prices = sim.read_model_and_scalers(model_path = model_path,
                                                                            media_scaler_path = media_scaler_path, 
                                                                            revenue_scaler_path = target_scaler_path)
 
########### FIND ALLOCATION ##############
########################################

budget = 2000
    
solution, kpi_without_optim, previous_budget_allocation = optimize_media.find_optimal_budgets(
                n_time_periods = 30,
                media_mix_model = mmm_model,
                budget = jnp.array(budget),
                prices = prices,
                media_scaler = media_scaler,
                target_scaler = target_scaler,
                seed = 52,
                bounds_lower_pct=0.2,
                bounds_upper_pct=0.2
    )
 
print(solution)

The first time I run the code with budget=2000, it runs properly. But If I change the budget parameter to budget=3000 and then run it again, I get a ConcretizationTypeError:  

"name": "ConcretizationTypeError", "message": "Abstract tracer value encountered where concrete value is expected: 
Traced<ShapedArray(bool[])>with<DynamicJaxprTrace(level=1/0)>\nThe problem arose with the `bool` function. \nThe error 
occurred while tracing the function _objective_function at 
.venv/lib/python3.9/site-packages/lightweight_mmm/optimize_media.py:27 for jit. This value became a tracer due to JAX 
operations on these lines:\n\n  operation a:f32[2] = copy b\n    from line 
.venv/lib/python3.9/site-packages/lightweight_mmm/lightweight_mmm.py:523 (predict)\n\n  operation a:f32[] b:f32[] = pjit[\n 
jaxpr={ lambda ; c:f32[2]. let\n      d:f32[1] = slice[limit_indices=(1,) start_indices=(0,) strides=(1,)] c\n      e:f32[] = 
squeeze[dimensions=(0,)] d\n      f:f32[1] = slice[limit_indices=(2,) start_indices=(1,) strides=(1,)] c\n      g:f32[] = 
squeeze[dimensions=(0,)] f\n    in (e, g) }\n  name=_unstack\n] h\n    from line 
.venv/lib/python3.9/site-packages/lightweight_mmm/lightweight_mmm.py:100 (<genexpr>)\n\n  operation a:bool[2] = eq b c\n  
from line .venv/lib/python3.9/site-packages/lightweight_mmm/lightweight_mmm.py:104 (_compare_equality_for_lmmm)\n\n 
operation a:bool[2] = pjit[\n  jaxpr={ lambda ; b:f32[2]. let c:bool[2] = ne b b in (c,) }\n  name=isnan\n] d\n    from line 
.venv/lib/python3.9/site-packages/lightweight_mmm/lightweight_mmm.py:104 (_compare_equality_for_lmmm)\n\n  operation 
a:bool[2] = pjit[\n  jaxpr={ lambda ; b:f32[2]. let c:bool[2] = ne b b in (c,) }\n  name=isnan\n] d\n    from line 
.venv/lib/python3.9/site-packages/lightweight_mmm/lightweight_mmm.py:104 (_compare_equality_for_lmmm)\n\nSee 
https://jax.readthedocs.io/en/latest/errors.html#jax.errors.ConcretizationTypeError"

I read the jax documentation trying to modify the package function but I was not able to fix this.    Can someone explain how I can handle this error ?   Thanks in advance !

alberto-molinaro commented 1 year ago

Hello everyone,

I found a fix: Downgrading jax from 0.4.8 to 0.4.2 fixed the error on mac M1.

Thank you!