`DiscreteDistribution.dist_of_func` about hark HOT 3 CLOSED

On Tue, Jun 20, 2023 at 7:22 PM Mateo Velásquez-Giraldo < ***@***.***> wrote: There is momentum gathering behind an xarray revolution <#1289>. After checking out the library and @alanlujan91 <https://github.com/alanlujan91> 's ConsLabeledModel <https://github.com/econ-ark/HARK/blob/37134b9c9eea4591659b7cc398cf5894e44ba0e7/HARK/ConsumptionSaving/ConsLabeledModel.py>, I am happy to join that revolution. A crucial part, I think, is getting the distribution machinery to work with x arrays. Which I think we do with the DiscreteDistributionLabeled class. I was playing around with its dist_of_func method and ran into behavior that is either unintuitive or a bug. Consider the following example code, which tries to take the expectation of a labeled function in two different ways. a) Directly, using the .expected method. b) First creating an object that represents the distribution of the function, and only then taking its expectation. # Create a basic labeled distribution base_dist = DiscreteDistributionLabeled( pmv=np.array([0.5, 0.5]), atoms=np.array([[1.0, 2.0], [3.0, 4.0]]), var_names=["a", "b"] ) # Define a transition function def transition(shocks, state): state_new = {} state_new["m"] = state["m"]*shocks["a"] state_new["n"] = state["n"]*shocks["b"] return state_new m = xr.DataArray(np.linspace(0, 10, 11), name="m", dims=("grid",)) n = xr.DataArray(np.linspace(0, -10, 11), name="n", dims=("grid",)) state_grid = xr.Dataset({"m": m, "n": n}) # Evaluate labeled transformation # This works new_state_exp = base_dist.expected(transition, state=state_grid) # This does not work new_state_dstn = base_dist.dist_of_func(transition, state=state_grid) new_state_dstn.expected() Method a) works but b) instead raises the error new_state_dstn.expected() File "/home/mvg/GitHub/HARK/HARK/distribution.py", line 1243, in expected return super().expected() File "/home/mvg/GitHub/HARK/HARK/distribution.py", line 936, in expected f_query = self.atoms AttributeError: 'DiscreteDistributionLabeled' object has no attribute 'atoms' The application of the function of the function is not yielding an object with an atoms dimension over which to integrate, I think. It might be that something else needs to be added to the function. But it is weird that .expected() can deal with it and .dist_of_func can not. I can look into it sometime, but wanted to confirm whether it is a bug, or a lack of documentation or simply a desired to-be-implemented feature. — Reply to this email directly, view it on GitHub <#1290>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAKCK75QN3FKELAMWYWORV3XMIWE7ANCNFSM6AAAAAAZN6HNLQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>