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

# # Spatial Autocorrelation

# * [Spatial Lag](#Spatial-Lag)
# * [Global autocorrelation](#Global-autocorrelation)
#     * [Moran's I](#Moran's-I)
#     * [Moran Plot](#Moran-Plot)
#     * [Inference](#Global-inference)
# * [Local autocorrelation](#Local-autocorrelation)
#     * [Local Moran's I](#Local-Moran's-I)
#     * [Clusters](#Clusters)
#     * [Inference](#Local-inference)
# * [Exercises](#Exercises)

# In[1]:


get_ipython().run_line_magic('matplotlib', 'inline')

import geopandas
import numpy
import pandas
import matplotlib.pyplot as plt
from pysal.lib import weights
from pysal.explore import esda
from pysal.viz.splot import esda as esdaplot

db = geopandas.read_file("../data/lux_regions.gpkg")
w = weights.Queen.from_dataframe(db)
w.transform = "R"


# ## Spatial Lag

# In[2]:


ll_lag = weights.lag_spatial(w, db["light_level"])
pandas.Series(ll_lag).head()


# ## Global autocorrelation

# ### Moran's I

# In[3]:


mi = esda.Moran(db["light_level"], w)


# ### Moran Plot

# In[4]:


esdaplot.moran_scatterplot(mi);


# ### Global inference

# In[5]:


esdaplot.plot_moran_simulation(mi, aspect_equal=False);


# ## Local autocorrelation

# ### Local Moran's I

# In[6]:


lisa = esda.Moran_Local(db["light_level"], w)


# ### Clusters

# In[7]:


esdaplot.lisa_cluster(lisa, db);


# * Putting everything together

# In[8]:


esdaplot.plot_local_autocorrelation(lisa, db, "light_level");


# ### Local inference

# In[9]:


pandas.Series(lisa.p_sim).head()


# In[10]:


f, axs = plt.subplots(1, 3, figsize=(16, 3))
for i, ax in zip([0.05, 0.01, 0.001], axs):
    esdaplot.lisa_cluster(lisa, db, p=i, ax=ax)
    ax.set_title(f"Confidence threshold = {i}")


# ## Exercises

# * Plot the distribution of values for `light_level` and its spatial lag. How do they relate to each other? What can you say about the lag in comparison to the original variable?
# * Compare the results presented above with a similar analysis of `POPULATION`. What are the main similarities and differences?