

In [ ]:

    
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np

Read data from the CSV file



In [ ]:

    
pk = pd.read_csv("../input/pokemon.csv")



In [ ]:

    
pk.head()



In [ ]:

    
pk.columns

Remove invalid data



In [ ]:

    
pk = pk[pk["capture_rate"] != "30 (Meteorite)255 (Core)"]
pk.capture_rate = pk.capture_rate.astype(np.int64)

Fill missing values



In [ ]:

    
pk["height_m"] = pk["height_m"].fillna(pk["height_m"].median())
pk["percentage_male"] = pk["percentage_male"].fillna(-1)
pk["weight_kg"] = pk["weight_kg"].fillna(pk["weight_kg"].median())



In [ ]:

    
pk.describe()

Influence of the Primary Type

This is the number of Pokemons for each type :



In [ ]:

    
pk["type1"].value_counts()

There are a lot more Water and Normal Pokemons than other types.

Influence of the Primary type on the capture rate



In [ ]:

    
capture_type = pk[["capture_rate", "type1"]]
capture_type = capture_type.groupby("type1").median().sort_values("capture_rate", ascending=False)
capture_type.plot(kind="bar", title="Capture rate for each (Primary) type")
plt.show()

Which type is best against which



In [ ]:

    
best_against = pk.groupby("type1").median()
best_against = best_against[['against_bug', 'against_dark', 'against_dragon',
       'against_electric', 'against_fairy', 'against_fight', 'against_fire',
       'against_flying', 'against_ghost', 'against_grass', 'against_ground',
       'against_ice', 'against_normal', 'against_poison', 'against_psychic',
       'against_rock', 'against_steel', 'against_water']].transpose()

best_against.style.highlight_max()



In [ ]:

    
best_against.sum().sort_values(ascending=False)

The Ice type seems to be the most vulnerable type, in part because Flying Pokemons are really powerful against Ice Pokemons.

But the type which has most enemies best against it is the Ground type.

How many Pokemon of each type are legendaries ?



In [ ]:

    
legendary_type = pk[["type1", "is_legendary"]]
legendaries = legendary_type[legendary_type["is_legendary"] == 1]



In [ ]:

    
chance_of_legend = legendaries.groupby("type1").count() / legendary_type.groupby("type1").count()
chance_of_legend = chance_of_legend.sort_values("is_legendary", ascending=False)
chance_of_legend.plot(kind="bar", title="Proportion of legendaries Pokemon of each (Primary) type")
plt.show()

1/3 of the Flying Pokemons are legendaries, whereas none of the Fighting and Poison types are legendaries.

Which type looks like what ?



In [ ]:

    
pokemons = pk.groupby("type1").median()
pokemons_height_weight = pokemons[["height_m", "weight_kg"]]

plt.rcParams['figure.figsize'] = [15, 8]

ax = pokemons.plot.scatter(x='weight_kg', y='height_m')
for i, txt in enumerate(pokemons.index):
    ax.annotate(txt, (pokemons.weight_kg.iat[i],pokemons.height_m.iat[i]))
plt.show()

Try to predict legendaries Pokemons

Preprocessing



In [ ]:

    
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn import preprocessing

def remove_unnecessary_features(df):
    del df["abilities"]
    del df["name"]
    del df["japanese_name"]
    del df["classfication"]
    del df["type2"]
    
def label_encode(df, column):
    label_encoder = preprocessing.LabelEncoder()
    label_encoder.fit(df[column])
    df[column] = label_encoder.transform(df[column])
    return label_encoder

X = pk.copy()
X.drop([25, 149, 150]) # Let's remove some Pokemons to try and predict them later
remove_unnecessary_features(X)
type_encoder = label_encode(X, "type1")

y = X.copy()["is_legendary"]
del X["is_legendary"]

Split train and test



In [ ]:

    
X_train, X_test, y_train, y_test = train_test_split(X, y)

Train

The random forest classifier is an ensemble learning method for classification that operates by constructing a multitude of decision trees at training time and outputting the class that is the mode of the classes (classification) of the individual trees. Random decision forests correct for decision trees' habit of overfitting to their training set.



In [ ]:

    
classifier = RandomForestClassifier(n_estimators=5)
classifier.fit(X_train, y_train)



In [ ]:

    
import seaborn as sns

cm = sns.light_palette("green", as_cmap=True)

feat_importance = pd.DataFrame([dict(zip(X_train.columns, classifier.feature_importances_))]).transpose().sort_values(0, ascending=False)
feat_importance.style.background_gradient(cmap=cm)

Test



In [ ]:

    
y_hat = classifier.predict(X_test)



In [ ]:

    
from sklearn.metrics import f1_score
f1_score(y_test, y_hat)

Predict

Mew and Mewtwo



In [ ]:

    
X_predict = pk.loc[[149, 150]]
remove_unnecessary_features(X_predict)
type_encoder = label_encode(X_predict, "type1")
del X_predict["is_legendary"]



In [ ]:

    
classifier.predict(X_predict)

Pikachu



In [ ]:

    
X_predict = pk.loc[[25]]
remove_unnecessary_features(X_predict)
type_encoder = label_encode(X_predict, "type1")
del X_predict["is_legendary"]
classifier.predict(X_predict)