Olivier Brunet

Data geek :)

9 minute read 

import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)

import pandas as pd

%matplotlib inline
import matplotlib.pyplot as plt
import seaborn as sns

SCRAPED_CSV = 'scraped_cars.csv'

Load CSV and Review

df_raw = pd.read_csv(SCRAPED_CSV)
df = df_raw.copy() # keep a defensive copy of the original data
df.tail()
name cylinders weight year territory acceleration mpg hp displacement
401 Ford Mustang Gl 4 2790 1982 USA 15.6 27.0 86.0 140.0
402 Vw Pickup 4 2130 1982 Europe 24.6 44.0 52.0 97.0
403 Dodge Rampage 4 2295 1982 USA 11.6 32.0 84.0 135.0
404 Ford Ranger 4 2625 1982 USA 18.6 28.0 79.0 120.0
405 Chevy S-10 4 2720 1982 USA 19.4 31.0 82.0 119.0 
df.shape

(406, 9)

df.sample(5)
name cylinders weight year territory acceleration mpg hp displacement
54 Pontiac Firebird 6 3282 1971 USA 15.0 19.0 100.0 250.0
313 Chevrolet Citation 6 2595 1979 USA 11.3 28.8 115.0 173.0
175 Ford Pinto 4 2639 1975 USA 17.0 23.0 83.0 140.0
75 Buick Lesabre Custom 8 4502 1972 USA 13.5 13.0 155.0 350.0
110 Chevrolet Impala 8 4997 1973 USA 14.0 11.0 150.0 400.0 
df.describe()
cylinders weight year acceleration mpg hp displacement
count 406.000000 406.000000 406.000000 406.000000 398.000000 400.000000 406.000000
mean 5.475369 2979.413793 1975.921182 15.519704 23.514573 105.082500 194.779557
std 1.712160 847.004328 3.748737 2.803359 7.815984 38.768779 104.922458
min 3.000000 1613.000000 1970.000000 8.000000 9.000000 46.000000 68.000000
25% 4.000000 2226.500000 1973.000000 13.700000 17.500000 75.750000 105.000000
50% 4.000000 2822.500000 1976.000000 15.500000 23.000000 95.000000 151.000000
75% 8.000000 3618.250000 1979.000000 17.175000 29.000000 130.000000 302.000000
max 8.000000 5140.000000 1982.000000 24.800000 46.600000 230.000000 455.000000

Data Review Strategy

df.territory.value_counts()

USA       254
Japan      79
Europe     73
Name: territory, dtype: int64

df.cylinders.value_counts()

4    207
8    108
6     84
3      4
5      3
Name: cylinders, dtype: int64

df.cylinders.value_counts().sort_index()

3      4
4    207
5      3
6     84
8    108
Name: cylinders, dtype: int64

ax = sns.countplot(data=df, y='cylinders')
ax.set_title("Counts for Each Cylinder Type");

png

ax = sns.countplot(data=df, x='year')
ax.set_title('Counts of Cars per Year');
plt.xticks(rotation=45);

png

df.isnull().sum()

name            0
cylinders       0
weight          0
year            0
territory       0
acceleration    0
mpg             8
hp              6
displacement    0
dtype: int64

df.mpg.isnull()[5:15]

5     False
6     False
7     False
8     False
9     False
10     True
11     True
12     True
13     True
14     True
Name: mpg, dtype: bool

df[df.mpg.isnull()]
name cylinders weight year territory acceleration mpg hp displacement
10 Citroen Ds-21 Pallas 4 3090 1970 Europe 17.5 NaN 115.0 133.0
11 Chevrolet Chevelle Concours (Sw) 8 4142 1970 USA 11.5 NaN 165.0 350.0
12 Ford Torino (Sw) 8 4034 1970 USA 11.0 NaN 153.0 351.0
13 Plymouth Satellite (Sw) 8 4166 1970 USA 10.5 NaN 175.0 383.0
14 Amc Rebel Sst (Sw) 8 3850 1970 USA 11.0 NaN 175.0 360.0
17 Ford Mustang Boss 302 8 3353 1970 USA 8.0 NaN 140.0 302.0
39 Volkswagen Super Beetle 117 4 1978 1971 Europe 20.0 NaN 48.0 97.0
367 Saab 900S 4 2800 1981 Europe 15.4 NaN 110.0 121.0 
df[df.hp.isnull()]
name cylinders weight year territory acceleration mpg hp displacement
38 Ford Pinto 4 2046 1971 USA 19.0 25.0 NaN 98.0
133 Ford Maverick 6 2875 1974 USA 17.0 21.0 NaN 200.0
337 Renault Lecar Deluxe 4 1835 1980 Europe 17.3 40.9 NaN 85.0
343 Ford Mustang Cobra 4 2905 1980 USA 14.3 23.6 NaN 140.0
361 Renault 18I 4 2320 1981 Europe 15.8 34.5 NaN 100.0
382 Amc Concord Dl 4 3035 1982 USA 20.5 23.0 NaN 151.0 
def highlight_max(s):
    '''Highlight the maximum in a Series yellow'''
    is_max = s == s.max()
    return ['background-color: yellow' if v else '' for v in is_max]

df.pivot_table(index=['year'], aggfunc='count').style.apply(highlight_max, axis=1)
acceleration cylinders displacement hp mpg name territory weight
year
1970 35 35 35 35 29 35 35 35
1971 29 29 29 28 28 29 29 29
1972 28 28 28 28 28 28 28 28
1973 40 40 40 40 40 40 40 40
1974 27 27 27 26 27 27 27 27
1975 30 30 30 30 30 30 30 30
1976 34 34 34 34 34 34 34 34
1977 28 28 28 28 28 28 28 28
1978 36 36 36 36 36 36 36 36
1979 29 29 29 29 29 29 29 29
1980 29 29 29 27 29 29 29 29
1981 30 30 30 29 29 30 30 30
1982 31 31 31 30 31 31 31 31 
print(f"Before we drop NaN rows we have {df.shape} rows")
df = df.dropna()
print(f"After we drop NaN rows we have {df.shape} rows")

Before we drop NaN rows we have (406, 9) rows
After we drop NaN rows we have (392, 9) rows

df.cylinders.value_counts().sort_index()

3      4
4    199
5      3
6     83
8    103
Name: cylinders, dtype: int64

df = df.query("cylinders != 3 and cylinders != 5").copy()
df.cylinders.value_counts().sort_index()

4    199
6     83
8    103
Name: cylinders, dtype: int64

df['cylinders_label'] = df.cylinders.apply(lambda v: f"{v} cylinders")
df.cylinders_label.value_counts().sort_index()

4 cylinders    199
6 cylinders     83
8 cylinders    103
Name: cylinders_label, dtype: int64

What Distribution Does MPG Have?

print(f"Examples of MPG values: {list(df.mpg.sample(10).sort_values())}")
ax = df.mpg.hist()
ax.set_ylabel('Frequency')
ax.set_xlabel('Binned MPG')
ax.set_title('Histogram of Continuous MPG Values');

Examples of MPG values: [14.0, 17.0, 18.0, 20.2, 23.5, 24.0, 25.0, 30.5, 36.0, 37.2]

png

ax = df.weight.hist()
ax.set_ylabel('Frequency')
ax.set_xlabel('Binned Weight')
ax.set_title('Histogram of Continuous Weight (lbs) Values');

png

Reviewing Our Goal

df.pivot_table(index=['year'], aggfunc='mean')
acceleration cylinders displacement hp mpg weight
year
1970 12.948276 6.758621 281.413793 147.827586 17.689655 3372.793103
1971 15.000000 5.629630 213.888889 107.037037 21.111111 3030.592593
1972 15.185185 5.925926 223.870370 121.037037 18.703704 3271.333333
1973 14.333333 6.461538 261.666667 131.512821 17.076923 3452.230769
1974 16.173077 5.230769 170.653846 94.230769 22.769231 2878.038462
1975 16.050000 5.600000 205.533333 101.066667 20.266667 3176.800000
1976 15.941176 5.647059 197.794118 101.117647 21.573529 3078.735294
1977 15.507407 5.555556 195.518519 104.888889 23.444444 3007.629630
1978 15.802857 5.371429 179.142857 99.600000 24.168571 2862.714286
1979 15.660714 5.857143 207.535714 102.071429 25.082143 3038.392857
1980 17.084000 4.160000 117.720000 77.000000 34.104000 2422.120000
1981 16.325000 4.642857 136.571429 81.035714 30.185714 2530.178571
1982 16.510000 4.200000 128.133333 81.466667 32.000000 2434.166667

What Correlates with MPG?

df.corr()['mpg'].sort_values()

weight         -0.842681
displacement   -0.817887
cylinders      -0.794872
hp             -0.780259
acceleration    0.419337
year            0.579778
mpg             1.000000
Name: mpg, dtype: float64

Exploring MPG vs. Weight

ax = df.plot(kind="scatter", x='weight', y='mpg')
ax.set_title('Scatterplot of MPG vs. Weight (lbs)');

png

ax = df.plot(kind="scatter", x='weight', y='mpg', alpha=0.5)
ax.set_title('Scatterplot of MPG vs. Weight (lbs)');

png

jg = sns.jointplot(data=df, x='weight', y='mpg');
jg.fig.suptitle('Somewhat Non-linear Relationship between HP and Weight');

png

jg=sns.jointplot(data=df, x='weight', y='mpg', kind='hexbin')
jg.fig.suptitle('Hexbin of Counts for MPG vs. Weight');

png

jg = sns.jointplot(data=df, x='weight', y='mpg').plot_joint(sns.kdeplot, zorder=0, n_levels=5)
jg.fig.suptitle('Density Estimate for MPG vs. Weight');

png

Exploring HP vs. Weight

ax = df.plot(kind="scatter", x='hp', y='weight')
ax.set_title('Reasonably Linear Relationship between HP and Weight');

png

jg = sns.jointplot(data=df, x='hp', y='weight', kind='reg')
jg.fig.suptitle('Reasonably Linear Relationship between HP and Weight');

png

df.query("hp > 200 and weight < 3500") # Fully equipped luxury entrant
name cylinders weight year territory acceleration mpg hp displacement cylinders_label
19 Buick Estate Wagon (Sw) 8 3086 1970 USA 10.0 14.0 225.0 455.0 8 cylinders

Cylinders and Displacement

sorted_cylinders_label = df.cylinders_label.value_counts().sort_index().index
sorted_cylinders_label

Index(['4 cylinders', '6 cylinders', '8 cylinders'], dtype='object')

ax = df.plot(kind='scatter', x='cylinders', y='displacement')
ax.set_title("Scatterplot of Cylinders vs. Displacement");

png

ax = sns.stripplot(data=df, x='cylinders_label', y='displacement', order=sorted_cylinders_label)
ax.set_title("Stripplot of Cylinders vs. Displacement with Jitter");

png

ax = sns.boxplot(data=df, x='cylinders_label', y='weight', order=sorted_cylinders_label)
ax.set_title("Boxplot of Cylinders vs. Weight");

png

ax = sns.boxplot(data=df, x='cylinders_label', y='displacement', order=sorted_cylinders_label, notch=True) 
ax.set_title("Notched Boxplot of Cylinders vs. Weight");

png

Looking at MPG over Time

ax = df.plot(kind="scatter", x='year', y='mpg');
ax.set_title('Scatterplot of MPG by Year');

png

ax = sns.stripplot(data=df, x='year', y='mpg')
ax.set_title('Stripplot of MPG by Year');
plt.xticks(rotation=45);

png

ax = sns.boxplot(data=df, x='year', y='mpg')
ax.set_title("Boxplot of MPG by Year");
plt.xticks(rotation=45);

png

df.query('year=="1978" and mpg > 40')
name cylinders weight year territory acceleration mpg hp displacement cylinders_label
251 Volkswagen Rabbit Custom Diesel 4 1985 1978 Europe 21.5 43.1 48.0 90.0 4 cylinders 
ax = sns.scatterplot(data=df, x='year', y='mpg', size='weight')
ax.set_title('Stripplot of MPG by Year');

png

ax = sns.scatterplot(data=df, x='year', y='mpg', hue='weight')
ax.set_title('Stripplot of MPG by Year');

png

ax = sns.boxplot(data=df, x='year', y='weight')
ax.set_title('Boxplot of Weight by Year');
plt.xticks(rotation=45);

png

Text analysis

df.head()
name cylinders weight year territory acceleration mpg hp displacement cylinders_label
0 Chevrolet Chevelle Malibu 8 3504 1970 USA 12.0 18.0 130.0 307.0 8 cylinders
1 Buick Skylark 320 8 3693 1970 USA 11.5 15.0 165.0 350.0 8 cylinders
2 Plymouth Satellite 8 3436 1970 USA 11.0 18.0 150.0 318.0 8 cylinders
3 Amc Rebel Sst 8 3433 1970 USA 12.0 16.0 150.0 304.0 8 cylinders
4 Ford Torino 8 3449 1970 USA 10.5 17.0 140.0 302.0 8 cylinders 
ser_car_makes = df.name.str.lower().str.split(n=1, expand=True)[0]
ser_car_makes.value_counts()[:10]

ford          48
chevrolet     43
plymouth      31
dodge         28
amc           27
toyota        25
datsun        23
buick         17
pontiac       16
volkswagen    15
Name: 0, dtype: int64

ser_car_makes.value_counts()[-10:]

mercedes-benz    2
cadillac         2
toyouta          1
triumph          1
maxda            1
chevroelt        1
capri            1
vokswagen        1
nissan           1
hi               1
Name: 0, dtype: int64

df['car_makes'] = ser_car_makes
ser_cars_by_territory = df[['territory', 'car_makes']].pivot_table(index=['territory', 'car_makes'], aggfunc='size')
df_cars_by_territory = pd.DataFrame(ser_cars_by_territory) # anonymous Series
df_cars_by_territory.columns = ['size'] # rename anonymous column 0 to a named column
df_cars_by_territory = df_cars_by_territory.sort_values(by=['territory', 'size'], ascending=False)
df_cars_by_territory
size
territory car_makes
USA ford 48
chevrolet 43
plymouth 31
dodge 28
amc 27
buick 17
pontiac 16
mercury 11
oldsmobile 10
chrysler 6
chevy 3
cadillac 2
capri 1
chevroelt 1
hi 1
Japan toyota 25
datsun 23
honda 13
mazda 7
subaru 4
maxda 1
nissan 1
toyouta 1
Europe volkswagen 15
fiat 8
peugeot 8
volvo 6
vw 6
audi 5
opel 4
saab 4
renault 3
bmw 2
mercedes-benz 2
triumph 1
vokswagen 1 
mask = df_cars_by_territory.apply(lambda x: x['size'] > 10, axis=1)
df_cars_by_territory[mask]
size
territory car_makes
USA ford 48
chevrolet 43
plymouth 31
dodge 28
amc 27
buick 17
pontiac 16
mercury 11
Japan toyota 25
datsun 23
honda 13
Europe volkswagen 15

Report

This report answers the following questions:

  • Is there a relationship between Weight and MPG - yes, with more weight we generally see lower MPG
  • Is there a relationship between Cylinders and Displacement - yes, with more Cylinders we see a higher Displacement
  • Is there a relationship between MPG and Years - yes, generally as we move from 1970 to 1982 we see steady improvements in MPG
print(f"df_raw had shape {df_raw.shape}, after dropping NaN entries we \
reduced from {df_raw.shape[0]} to {df.shape[0]} rows.")

assert ((df.mpg >= 7) & (df.mpg <=47)).all(), "Why do we see a wider range of MPG values now?"
assert ((df.year >= 1970) & (df.year <= 1982)).all(), "Why do we see a wider range of years now?"

df_raw had shape (406, 9), after dropping NaN entries we reduced from 406 to 385 rows.

As weight increases we see a decrease in MPG, this relationship is slightly non-linear with a faster decrease in MPG associated with low-to-mid Weight.

jg = sns.jointplot(data=df, x='weight', y='mpg').plot_joint(sns.kdeplot, zorder=0, n_levels=5)
jg.fig.suptitle('Density Estimate for MPG vs. Weight');

png

Acceleration gets worse as Horsepower decreases.

jg = sns.jointplot(data=df, x='acceleration', y='hp', kind='reg')
jg.fig.suptitle('Regression plot for Acceleration vs. Horsepower');

png

An engine’s cubic inches of displacement is strongly related to the count of Cylinders.

ax = sns.boxplot(data=df, x='cylinders_label', y='displacement', order=sorted_cylinders_label, notch=True) 
sns.stripplot(data=df, x='cylinders_label', y='displacement', order=sorted_cylinders_label, alpha=0.5, ax=ax) 
ax.set_title("More Cylinders Generally Have\nHigher Cubic Inches of Displacement");

png

Heavier vehicles tend to have more Horsepower, the heaviest vehicles have the highest count of Cylinders.

fig, ax = plt.subplots(figsize=(6, 6))
sns.scatterplot(data=df, x='weight', y='hp', hue='cylinders_label', 
                     hue_order=sorted_cylinders_label, ax=ax)
ax.set_title('Scatterplot of HP vs. Weight\nColoured by Cylinder Count');

png

Vehicle brands are strongly associated with Territory.

cm = sns.light_palette("green", as_cmap=True)
mask = df_cars_by_territory.apply(lambda x: x['size'] > 10, axis=1)
df_cars_by_territory[mask].style.background_gradient(cmap=cm)
size
territory car_makes
USA ford 48
chevrolet 43
plymouth 31
dodge 28
amc 27
buick 17
pontiac 16
mercury 11
Japan toyota 25
datsun 23
honda 13
Europe volkswagen 15

Vehicles from the USA consistently have worse MPG than corresponding vehicles from Japan or Europe in this sample. Vehicles in each territory become more efficient over the Years.

fg = sns.lmplot(data=df, x='year', y='mpg', hue='territory')
fg.fig.suptitle('MPG over Year by Territory');

png

You may also enjoy

H&M Personalized Fashion 1/2 - EDA

11 minute read

Exploratory analysis of articles, customers, and transactions datasets with recommendation engine specific caracteristics, such as a long tail.

What is federated learning?

10 minute read

A sub-field of ML focusing on decentralization, motivated by issues such as data privacy, data minimization, and data access rights.