%matplotlib inline
import metapack as mp
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import geopandas as gpd 

regions = mp.open_package('http://library.metatab.org/sandiego.gov-police_regions-1.csv')
regions

calls_p = mp.open_package('http://library.metatab.org/sandiego.gov-police_calls-2015e-1.csv')
calls_p

calls_r = calls_p.resource('pd_calls')
calls_r

call_type_r = calls_p.resource('call_type')
call_types = call_type_r.dataframe().rename(columns={'description':'call_type_desc'})
call_types.head()

regions_r = regions.resource('pd_beats')
regions_r

# The beats.cx[:-116.8,:] bit indexes the bounding box to exclude the empty portion of the 
# county. San Diego owns the footprint of a dam in east county, which displays as a tiny
# dot in the middle of empty space. 
# Note that this isn't actually defininf the bounding box; it's cutting out far-east regions, 
# and then GeoPandas creates the smaller bounding box that excludes them. So, the actually 
# value in the cx indexder can vary a bit. 

# Converting to float makes merging with the calls df ewasier, since the beat column
# in that df has nans. 

beats = regions_r.dataframe().geo
beats['beat'] = beats.beat.astype(float)
beats = beats.set_index('beat').cx[:-116.55,:]
beats.plot();

pd_calls = calls_r.read_csv(low_memory=False)

def augment_time(df):
    df['date_time'] = pd.to_datetime(df.date_time)
    
    df['hour'] = df.date_time.dt.hour
    df['month'] = df.date_time.dt.month
    df['year'] = df.date_time.dt.year
    df['dayofweek'] = df.date_time.dt.dayofweek
    df['weekofyear'] = df.date_time.dt.weekofyear

    df['weekofdata'] = (df.year-df.year.min())*52+df.date_time.dt.weekofyear
    df['monthofdata'] = (df.year-df.year.min())*12+df.date_time.dt.month

    return df

assert pd_calls.call_type.dtype == call_types.call_type.dtype

pd_calls = augment_time(pd_calls).merge(call_types, on='call_type')

pd_calls['beat'] = pd_calls.beat.astype(float)

pd_calls = pd_calls.merge(beats.reset_index()[['beat', 'name']], on='beat')\
    .rename(columns={'name':'beat_name'})

def plot_geo(df, color_col, title=None):
    # Need to get aspect right or it looks wacky
    bb = beats.total_bounds
    aspect = (bb[3] - bb[1])/ (bb[2]-bb[0])
    x_dim = 8

    fig = plt.figure(figsize = (x_dim,x_dim*aspect))
    ax = fig.add_subplot(111)

    df.plot(ax=ax,column=color_col, cmap='RdYlGn_r',
                     scheme='fisher_jenks', legend=True);
  
    if title:
        fig.suptitle(title, fontsize=18);

    leg = ax.get_legend()
   
    #leg.set_bbox_to_anchor((0., 1.02, 1., .102))
    leg.set_bbox_to_anchor((1,.5))

    plt.tight_layout(rect=[0, 0.03, 1, 0.95])
        
_ = gpd.GeoDataFrame(pd_calls.groupby('beat').incident_num.count().to_frame()\
                               .join(beats))

plot_geo(_, 'incident_num', 'Incidents Per Beat, 2015 to Aug 2017')

pd_calls.call_type_desc.value_counts().iloc[10:30]

_ = gpd.GeoDataFrame(pd_calls[pd_calls.call_type_desc == 'LOUD PARTY']
                             .groupby('beat')
                             .incident_num.count().to_frame()\
                             .join(beats))

plot_geo(_, 'incident_num', "LOUD PARTY calls, 2015 to Aug 2017")             

_ = gpd.GeoDataFrame(pd_calls[pd_calls.call_type_desc == 'ILLEGAL PARKING']
                             .groupby('beat')
                             .incident_num.count().to_frame()\
                             .join(beats))

plot_geo(_, 'incident_num',  "ILLEGAL PARKING calls, 2015 to Aug 2017")  

_ = pd_calls[pd_calls.call_type_desc == 'BATTERY']\
                             .groupby('beat')\
                             .incident_num.count().to_frame()\
                             .join(beats)


plot_geo(gpd.GeoDataFrame(_), 'incident_num', "BATTERY calls, 2015 to Aug 2017")  

_.sort_values('incident_num', ascending=False).head(10)

# Could also get the beats by name. 
pb_beat = beats[beats.name=='PACIFIC BEACH'].index.values[0]
gas_beat = beats[beats.name=='GASLAMP'].index.values[0]

low_crime = _.sort_values('incident_num', ascending=False).iloc[5:]
_lc = _.loc[list(low_crime.index.values)]

plot_geo(gpd.GeoDataFrame(_lc), 'incident_num', 
         "BATTERY calls, 2015 to Aug 2017, Lower Crime Areas")  

_ = gpd.GeoDataFrame(pd_calls[pd_calls.call_type_desc == 'BATTERY']
                             .groupby('beat')
                             .incident_num.count().to_frame()\
                             .join(beats))

plot_geo(_, 'incident_num', "BATTERY calls, 2015 to Nov 2017")  

_ = gpd.GeoDataFrame(pd_calls[pd_calls.call_type_desc == 'MENTAL CASE']
                             .groupby('beat')
                             .incident_num.count().to_frame()\
                             .join(beats))

plot_geo(_, 'incident_num', "MENTAL CASE calls, 2015 to Aug 2017")  

pb_beat = beats[beats.name=='PACIFIC BEACH'].index.values[0]

_ = pd_calls[(pd_calls.call_type_desc=='LOUD PARTY') & (pd_calls.beat == pb_beat)]

ht = pd.pivot_table(data=_, 
                    values='incident_num', index=['hour'],columns=['dayofweek'],
                    aggfunc='count')

fig, ax = plt.subplots(figsize=(6,6))

sns.heatmap(ht, ax=ax);

pb_beat = beats[beats.name=='PACIFIC BEACH'].index.values[0]

_ = pd_calls[(pd_calls.call_type_desc=='LOUD PARTY') & (pd_calls.beat == pb_beat)]



fig, ax = plt.subplots(figsize=(8,8))

fig.suptitle("LOUD PARTY Calls In Pacific Beach\n2015 to Aug 2017\nBy Hour and Month", fontsize=18);

sns.heatmap(ht, ax=ax);

hm_beats = pd_calls[['beat_name', 'hour','month']].copy()
hm_beats['count'] = 1
hm_beats = hm_beats.groupby(['beat_name', 'hour','month']).count().reset_index()

# Top 16 beats
top_beats= pd_calls.beat_name.value_counts().index.values[:16]

from IPython.display import display

# select only the rows for the top 16 beats
_ = hm_beats[hm_beats.beat_name.isin(top_beats)]

g = sns.FacetGrid(_, col="beat_name", col_wrap=4)

def facet_heatmap(data,  color, **kwargs):

    ht = data.pivot(index="hour", columns='month', values='count')

    sns.heatmap(ht, cmap='Reds', **kwargs)  

#cbar_ax = g.fig.add_axes([.92, .3, .02, .4])  # Create a colorbar axes

    
with sns.plotting_context(font_scale=3.5):   
    g = g.map_dataframe(facet_heatmap) 
    
plt.tight_layout(rect=[0, 0.03, 1, 0.95])
g.fig.suptitle("LOUD PARTY Calls By Month of Year, By Hour of Day, By Beat",
              fontsize=18);