Description

As with any good EDA, you try to understand the variables included.
i). What is the granularity of the data (i.e. what does each row represent)?
ii). As we discussed in class, classifications of variable conceptual types can sometimes be subjective depending on what we are doing with the dataset. Categorize each of the variables in this dataset as either
A). Quantitative: Continuous
B). Quantitative: Discrete
C). Categorical/Qualitative: Nominal
D). Categorical/Qualitative: Ordinal
Give your answer as a table in the following form:
Column Name Category Explanation/Reasoning
CASENO category letter here reasoning here
cont’d … … …
Column Name Category Explanation/Reasoning
CASENO C This is because the data can be catgorized through crime type, but not ranked.
OFFENSE C This is because, crime is random, and cannot be ordinal, but nominal
EVENTDT B This is Discrete as it counts to the day to the hour
EVENTTM A This is because time is continous, and doesn’t stop. Furthermore it is consistent in increments.
CVLEGEND D As this shows what kind of crime they committed, we can rank them when categorized for the type of crime.
i) Each row represents each case that had been called into to Berkeley Police Department.

Column Name Category Explanation/Reasoning
CVDOW B As this is countable, we can set this as quantitative discrete
InDbDate B This has days of the week, and then to the hour.
Block_Location C The location of the crime occured is random
BLKADDA C Each crime occured is random
City C No order, but within the City of Berkeley, thus not ranked
State C No order, but is within City of
Berkely, which is within
California. Thus it is nominal
2

0.1 Question 2c
Use seaborn to create a density histogram showing the distribution of calls by hour. Include the Kernal Density Estimate (KDE) graph on your histogram.
Be sure that your axes are labeled and that your plot is titled.
In [20]: ax_3d = sns.histplot(data = calls, x = “Hour”, stat = ‘density’, binwidth = 0.999999, kde=True) …
# Your code above this line plt.xlabel(“Hour”) plt.ylabel(“Calls by Percentage”) plt.title(‘Histogram of Calls by Hour’) # Leave this for grading purposes ax_3d = plt.gca()

4

0.1.1 Question 2e
i). Use seaborn to construct a box plot showing the distribution of calls by hour.
ii). To better understand the time of day a report occurs we could stratify the analysis by DayType (i.e. by weekday vs weekends).
Use seaborn to create side-by-side violin plots comparing the distribution of calls by hour on the weekend vs weekday (hint: see the violin plot documentation on how to stratify by a column in the dataframe https://seaborn.pydata.org/generated/seaborn.violinplot.html )
Note: For aesthetic purposes only the violin plot continues past the end of the whiskers (i.e. past 0 and 24 hours); however it is not possible to get data points outside of the whiskers for this distribution.
In [23]: ax_3d = sns.boxplot(data = calls, x = “Hour”) plt.xlabel(“Hour”) plt.ylabel(“Calls”) plt.title(“Calls by the Hour”) # Your code for boxplot above this line
Out[23]: Text(0.5, 1.0, ‘Calls by the Hour’)

In [24]: ax_3d = sns.violinplot(data = calls, x = “Hour”, y = “DayType”) plt.xlabel(“Hour”) plt.ylabel(“DayType”) plt.title(“Calls by the Hour”)
# Your code for boxplot above this line# Your code for side-by-side violin plots above this lin Out[24]: Text(0.5, 1.0, ‘Calls by the Hour’)
6

8

0.2 Question 2f
Based on your histogram, boxplot, and violin plots above, what observations can you make about the patterns of calls? Answer each of the following questions:
• Are there more calls in the day or at night?
• What are the most and least popular times?
• Do call patterns and/or IQR vary by weekend vs weekday?
There are more calls in the afternoon around 3 to 8. The most times in both weekdays and weekends are around 3:00pm to 8:00 pm. The patterns seem the same, except during the weekend, there is a slight decrease in calls around 7am to 9am, creating a narrower violin graph.
10 In [34]: missing_by_crime = missing_lat_lon[“CVLEGEND”].value_counts() nomiss = calls[“CVLEGEND”].value_counts() # Your code above this line
missing_by_crime = ((missing_by_crime/(nomiss-missing_by_crime))
.dropna().sort_values(ascending = False))
In [35]: plt.barh(missing_by_crime.index, missing_by_crime) plt.xlabel(“Percentage of Crimes That Have No Valid Coordinates”)
# Your code to create the barplot above this line
Out[35]: Text(0.5, 0, ‘Percentage of Crimes That Have No Valid Coordinates’)

12
0.2.1 Question 3d
Based on the plots above, are there any patterns among entries that are missing latitude/longitude data?
Based on the plots above, give your recommendation as to how we should handle the missing data, and justify your answer:
Option 1). Drop rows with missing data
Option 2). Set missing data to NaN
Option 3). Impute data
14
0.3 Question 3e
Based on the above map, what could be some drawbacks of using the location fields in this dataset to draw conclusions about crime in Berkeley? Here are some sub-questions to consider:
• Zoom into the map. Why are all the calls located on the street and often at intersections?
16