#!/usr/bin/env python
# coding: utf-8

# # Refutation Notebook
# The objective of this notebook is to compare the ability of refuters to detect the problems in a given set of estimators.
# Note:
# This notebook makes use of the optional dependencies:
# - pygraphviz
# - causalml

# ## Import Dependencies

# In[1]:


from dowhy.datasets import linear_dataset
from dowhy import CausalModel
import causalml


# ## Inspection Parameters
# These parameters give us the option of inspecting the intermediate steps to sanity check the steps performed

# In[2]:


inspect_datasets = True
inspect_models = True
inspect_identified_estimands = True
inspect_estimates = True
inspect_refutations = True


# ## Estimator List
# We pass a list of strings, corresponding to the estimators of interest

# In[3]:


estimator_list = ["backdoor.propensity_score_matching", "backdoor.propensity_score_weighting", "backdoor.causalml.inference.meta.LRSRegressor"]
method_params= [ None, None, { "init_params":{} } ]


# ## Refuter List
# A list of strings, corresponding to each refuter we wish to run

# In[4]:


refuter_list = ["bootstrap_refuter", "data_subset_refuter"]


# ## Create the Datasets

# In[5]:


# Parameters for creating the Dataset
TREATMENT_IS_BINARY = True
BETA = 10
NUM_SAMPLES = 1000
NUM_CONFOUNDERS = 5
NUM_INSTRUMENTS = 3
NUM_EFFECT_MODIFIERS = 2

# Creating a Linear Dataset with the given parameters
linear_data = linear_dataset(
            beta = BETA,
            num_common_causes = NUM_CONFOUNDERS,
            num_instruments = NUM_INSTRUMENTS,
            num_effect_modifiers = NUM_EFFECT_MODIFIERS,
            num_samples = NUM_SAMPLES,
            treatment_is_binary = True
        )
# Other datasets come here 


# Append them together in an array
datasets = [linear_data]


# ## Inspect Data

# In[6]:


dataset_num = 1
if inspect_datasets is True:
    for data in datasets:
        print("####### Dataset {}###########################################################################################".format(dataset_num))
        print(data['df'].head())
        print("#############################################################################################################")
        dataset_num += 1


# ## Create the CausalModels

# In[7]:


models = []
for data in datasets:
    model = CausalModel(
                data = data['df'],
                treatment = data['treatment_name'],
                outcome = data['outcome_name'],
                graph = data['gml_graph']
            )
    models.append(model)


# ## Inspect Models

# In[8]:


model_num = 1
if inspect_models is True:
    for model in models:
        print("####### Model {}#############################################################################################".format(model_num))
        print("Common Causes:",model._common_causes)
        print("Effect Modifiers:",model._effect_modifiers)
        print("Instruments:",model._instruments)
        print("Outcome:",model._outcome)
        print("Treatment:",model._treatment)
        print("#############################################################################################################")
        model_num += 1


# ## Identify Effect

# In[9]:


identified_estimands = []
for model in models:
    identified_estimand = model.identify_effect(proceed_when_unidentifiable=True)
    identified_estimands.append(identified_estimand)


# ## Identified Estimands

# In[10]:


estimand_count = 1
for estimand in identified_estimands:
    print("####### Identified Estimand {}#####################################################################################".format(estimand_count))
    print(estimand)
    print("###################################################################################################################")
    estimand_count += 1


# ## Estimate Effect

# In[11]:


estimate_list = []
for i in range(len(identified_estimands)):
    for j in range(len(estimator_list)):
        estimate = model.estimate_effect(
                        identified_estimands[i],
                        method_name=estimator_list[j],
                        method_params=method_params[j]
                  )
        estimate_list.append(estimate)


# ## Estimate Values

# In[12]:


estimand_count = 1
if inspect_estimates is True:
    for estimand in estimate_list:
        print("####### Estimand {}#######################################################################################".format(estimand_count))
        print("*** Class Name ***")
        print()
        print(estimand.params['estimator_class'])
        print()
        print(estimand)
        print("########################################################################################################")
        print()
        estimand_count += 1
    


# ## Refute Estimate

# In[13]:


refutation_list = []
for estimand in identified_estimands:
    for estimate in estimate_list: 
        for refuter in refuter_list:
            ref = model.refute_estimate(estimand, estimate,method_name=refuter)
            refutation_list.append(ref)


# ## Refutation Values

# In[14]:


refuter_count = 1
if inspect_refutations is True:
    for refutation in refutation_list:
        print("####### Refutation {}#######################################################################################".format(refuter_count))
        print("*** Class Name ***")
        print()
        print(refutation.refutation_type)
        print()
        print(refutation)
        print("########################################################################################################")
        print()
        refuter_count += 1


# In[ ]:





# In[ ]: