search

DataCanvasIO / YLearn

YLearn, a pun of "learn why", is a python package for causal inference
https://ylearn.readthedocs.io
Apache License 2.0
391 stars 75 forks source link

decide attributeis not found #44

Closed yanduoduan closed 1 year ago

yanduoduan commented 1 year ago

AttributeError: 'PolicyInterpreter' object has no attribute 'decide'，why？

lixfz commented 1 year ago

PolicyInterpreter.decide() was added in the main branch which is not released yet. Upgrade YLearn from latest source code to run latest case study please.

yanduoduan commented 1 year ago

hi,There is one control and several treatment, I want to build the model through casual tree, the code shows an error, what is the problem? x = [[0,0.31388521, -1.0856306], [1,0.35388521, -11.0856306], [2,1.715597, -1.03306], [1,1.840204, -1.0506], [2,1.144243, -1.0876], [0,0.282745, -1.0806], [0,0.31388521, -4.046], [0,0.21388521, -1.3997345], [1,0.11388521, -2.0856306], [2,0.1388521, -1.0856306]] outcome = 'outcome' treatment = 'treatment' data = pd.DataFrame(x,columns=['treatment','outcome','v_0']) adjustment = data.columns[2:] ct = CausalTree(min_samples_leaf=3, max_depth=5) ct.fit(data=data, outcome=outcome, treatment=treatment, adjustment=adjustment,treat=[1,2],control=0)

BochenLv commented 1 year ago

Hi,

The error is due to that the parameter treat is given a wrong value [1, 2] when calling fit(). It should be either treat=1 or treat=2.

I further explain this issue as follows.

Curretnly CausalTree only supports binary treatment problems. That means when a treatment variable (call it T) has several different possible values (e.g., T=0, 1, 2), CausalTree will first convert this problem into a binary one, e.g., T=0 is the control while T=2 is the treat by specifiying the parameters treat=2 and control=0 when calling the fit() method. The fitted CausalTree will be able to estimate the causal effects of T=2 against T=0.

In this sense, I think your question is actually "There is one control and several treatment values for a single treatment variable".

When should one set the parameter treat as a list of different values? The answer is when there are several treatment variables. For an example, there are two treatment variables T_1 with possible values 0, 1, 2, 3 and T_2 with possible values 0, 1. Suppose that we take those data with T_1=0 and T_2=0 as the control group while take those data with T_1=3 and T_2=1 as the treat group, then when calling the fit() method we should set treat=[3, 1] and control=[0, 0].

yanduoduan commented 1 year ago

hi,I understand the problem.If my example above does not specify treat and control, how the causal tree is handled?

BochenLv commented 1 year ago

If the treat and control are not given, then CausalTree will first apply an OrdinalEncoder transformer to the treatment variable which will be transformed into an array of integers. After that, CausalTree will implicitly set treat as 1 and control as 0.

That said, treat=1 and control=0 implicitly in your example if they are not specified.

BochenLv commented 1 year ago

Here we assume that the OrdinalEncoder will transform 1 into 1 and 0 into 0.