Pasted

# %%
import numpy as np
import pandas as pd
from matplotlib import pyplot as plt
import lightgbm as lgb
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from sklearn.tree import DecisionTreeClassifier, export_graphviz
 
# %%
df = pd.read_csv('data/cars-train.csv')
# %%
 
 
def get(d, **kwargs):
    selector = np.ones(d.shape[0], dtype=bool)
    for name, value in kwargs.items():
        selector &= (d[name] == value)
    return d[selector]
 
 
def car(d):
    return get(d, car_or_bus='car')
 
 
def bus(d):
    return get(d, car_or_bus='bus')
 
 
# %%
car(df)['rating'].plot.hist(color='r', alpha=0.5, label='car')
bus(df)['rating'].plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Rating")
plt.legend()
plt.show()
# %%
car(df)['v'].plot.hist(color='r', alpha=0.5, label='car')
bus(df)['v'].plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Velocity")
plt.legend()
plt.show()
# %%
car(df)['t'].plot.hist(color='r', alpha=0.5, label='car')
bus(df)['t'].plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Time")
plt.legend()
plt.show()
# %%
car(df)['distance'].plot.hist(color='r', alpha=0.5, label='car')
bus(df)['distance'].plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Distance")
plt.legend()
plt.show()
# %%
(car(df)['v'] * car(df)['distance']).plot.hist(color='r', alpha=0.5, label='car')
(bus(df)['v'] * bus(df)['distance']).plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Velocity * Distance")
plt.legend()
plt.show()
# %%
(car(df)['t'] * car(df)['distance']).plot.hist(color='r', alpha=0.5, label='car')
(bus(df)['t'] * bus(df)['distance']).plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Time * Distance")
plt.legend()
plt.show()
# %%
(car(df)['distance'] * car(df)['distance']).plot.hist(color='r', alpha=0.5, label='car')
(bus(df)['distance'] * bus(df)['distance']).plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Distance * Distance")
plt.legend()
plt.show()
# %%
df['distance_sq'] = df['distance'] * df['distance']
car(df[df['distance_sq'] < 70])['distance_sq'].plot.hist(color='r', alpha=0.5, label='car')
bus(df[df['distance_sq'] < 70])['distance_sq'].plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Distance * Distance")
plt.legend()
plt.show()
# %%
df['t_sq'] = df['t'] * df['t']
car(df[df['t_sq'] < 0.5])['t_sq'].plot.hist(color='r', alpha=0.5, label='car')
bus(df[df['t_sq'] < 0.5])['t_sq'].plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Time * Time")
plt.legend()
plt.show()
# %%
df['v_sq'] = df['v'] * df['v']
car(df[df['v_sq'] < 200])['v_sq'].plot.hist(color='r', alpha=0.5, label='car')
bus(df[df['v_sq'] < 200])['v_sq'].plot.hist(color='b', alpha=0.5, label='bus')
plt.title("Velocity * Velocity")
plt.legend()
plt.show()
# %% Drop outliers
df.drop([133], inplace=True)  # Car with rating=1
# %%
df['car_or_bus'] = df['car_or_bus'].map({'car': 0, 'bus': 1})
# %%
X_train, X_valid, y_train, y_valid = train_test_split(df.drop('car_or_bus', axis=1), df['car_or_bus'], test_size=.2)
# %%
 
 
def acc(pred, true_data):
    return "accuracy", accuracy_score(true_data.label, pred > 0.5), True
 
 
# %%
# params = {
#     'objective': 'binary',
#     'num_iterations': 100,
#     'learning_rate': 0.01,
#     'num_leaves': 16,
#     'metric': 'auc',
#     'early_stopping_rounds': 5,
#     'seed': 6741
# }
# # train = lgb.Dataset(df.drop('car_or_bus', axis=1), df['car_or_bus'])
# train = lgb.Dataset(X_train[['rating']], y_train)
# valid = lgb.Dataset(X_valid[['rating']], y_valid, reference=train)
# booster = lgb.train(params, train_set=train, valid_sets=valid)
# %%
tree = DecisionTreeClassifier(min_samples_leaf=10)
tree.fit(df.drop('car_or_bus', axis=1), df['car_or_bus'])
# %%
split_data = {}
for s in np.linspace(0, 1, 10000):
    split_data[s] = accuracy_score(y_valid, (tree.predict_proba(X_valid) > s)[:, 1])
print(np.unique(list(split_data.values())))
print(max(split_data.items(), key=lambda x: x[1]))
# %%
test = pd.read_csv('data/cars-test.csv')
# %%
test['distance_sq'] = test['distance'] * test['distance']
test['v_sq'] = test['v'] * test['v']
test['t_sq'] = test['t'] * test['t']
# %%
pred = tree.predict_proba(test)[:, 1] > 0.36363636363636365
pred = pd.Series(pred).map({False: 'car', True: 'bus'})
pred.to_csv('subm.csv', index=False)