def process_trip(x, start_time):
tt = time.localtime(start_time)
data = [tt.tm_wday, tt.tm_hour]
# distance from the center till cutting point
v_mn = 0
head = 0
if len(x)>1:
v_mn = haversineKaggle(x[0,:], x[1,:])[0]
head = heading(x[0,:], x[1,:])
# distance from the center till cutting point
d_st = haversineKaggle(x[0,:], CITY_CENTER)
h_st = heading(x[0,:], CITY_CENTER[0])
data += [x[-1,0], x[-1,1], d_st, h_st, v_mn, head]
return data
def process_trip(x, start_time):
tt = time.localtime(start_time)
data = [tt.tm_wday, tt.tm_hour]
# distance from the center till cutting point
v_mn = 0
head = 0
if len(x) > 1:
v_mn = haversineKaggle(x[0, :], x[-1, :])[0]
head = heading(x[0, :], x[-1, :])
# distance from the center till cutting point
d_st = haversineKaggle(x[0, :], CITY_CENTER)
h_st = heading(x[0, :], CITY_CENTER[0])
data += [x[-1, 0], x[-1, 1], d_st, h_st, v_mn, head]
return data
def process_trip(x, start_time):
tt = time.localtime(start_time)
data = [tt.tm_wday, tt.tm_hour]
# distance from the center till cutting point
d_st = haversineKaggle(x, CITY_CENTER)
head = heading(x, CITY_CENTER[0])
data += [x[0], x[1], d_st, head]
return data
def process_trip(x, start_time):
tt = time.localtime(start_time)
data = [tt.tm_wday, tt.tm_hour]
# distance from the center till cutting point
d_st = haversineKaggle(x, CITY_CENTER)
head = heading(x, CITY_CENTER[0])
data += [x[0], x[1], d_st, head]
return data
def process_trip(x, start_time):
tt = time.localtime(start_time)
data = [tt.tm_wday, tt.tm_hour]
# cumulative sum of distance
d_cs = 0
vcar = 0
vmed = 0
head = 0
if x.shape[0] > 1:
d1 = haversineKaggle(x[:-1,:], x[1:,:])
d_cs = np.sum(d1)
vmed = np.median(d1)
vcar = d1[-1]
head = heading(x[-2,:], x[-1,:])
# distance from the center till cutting point
d_st = haversineKaggle(x[0,:], CITY_CENTER)[0]
h_st = heading(x[0,:], CITY_CENTER[0])
d_cut = haversineKaggle(x[-1,:], CITY_CENTER)[0]
h_cut = heading(CITY_CENTER[0], x[-1,:])
data += [x.shape[0], x[0,0], x[0,1], x[-1,0], x[-1,1], d_st, h_st, d_cut,
h_cut, d_cs, vmed, vcar, head]
return data
import numpy as np
import pandas as pd
from sklearn.ensemble import RandomForestRegressor
from utils import haversineKaggle
DATA_DIR = '../data'
t0 = time.time()
for filename in ['train_pp_N1.csv', 'train_pp_N2.csv', 'train_pp_N3.csv',
'train_pp_RND.csv']:
print('reading training data from %s ...' % filename)
df = pd.read_csv(os.path.join(DATA_DIR, filename))
d1 = haversineKaggle(df[['xs', 'ys']].values, df[['xe', 'ye']].values)
# create training set
y = np.log(df['len']*15 + 1)
# remove non-predictive features
df.drop(['CALL_TYPE', 'TAXI_ID', 'xe', 'ye', 'len'], axis=1, inplace=True)
X = np.array(df, dtype=np.float)
# clean data by removing long distance tracks
th1 = np.percentile(d1, [99.9])
X = X[(d1<th1), :]
y = y[(d1<th1)]
print('training a random forest regressor ...')
# Initialize the famous Random Forest Regressor from scikit-learn
clf = RandomForestRegressor(n_estimators=200, n_jobs=-1, random_state=21)
if not os.path.isfile(filename):
continue
df = pd.read_csv(filename)
if df.shape[0] < 1000:
print('skipping key point %i (%i)' % (id_, df.shape[0]))
continue
# factorize categorical columns in training set
#df['CALL_TYPE'], ct_index = pd.factorize(df['CALL_TYPE'])
#df = df[df['CALL_TYPE'] == 0] # A=2, B=1, C=0
# fill all NaN values with -1
#df = df.fillna(-1)
# remove long distance
d1 = haversineKaggle(df[['xs', 'ys']], df[['xe', 'ye']])
th1 = np.percentile(d1, [99.9])
df = df.loc[d1 < th1]
y = np.ravel(np.log(df['len'] * 15 + 1))
df.drop(['CALL_TYPE', 'TAXI_ID', 'xe', 'ye', 'len'], axis=1, inplace=True)
X = np.array(df, dtype=np.float)
print('training classifier of key point %i (sz=%i) ...' %
(id_, X.shape[0]))
# Initialize the famous Random Forest Regressor from scikit-learn
clf = RandomForestRegressor(n_estimators=200, n_jobs=-1, random_state=21)
clf.fit(X, y)
pred_rf = clf.predict(X_tst[id_, :])
clf = GradientBoostingRegressor(n_estimators=200,
import pandas as pd
from sklearn.ensemble import RandomForestRegressor
from utils import haversineKaggle
DATA_DIR = '../data'
t0 = time.time()
for filename in [
'train_pp_N1.csv', 'train_pp_N2.csv', 'train_pp_N3.csv',
'train_pp_RND.csv'
]:
print('reading training data from %s ...' % filename)
df = pd.read_csv(os.path.join(DATA_DIR, filename))
d1 = haversineKaggle(df[['xs', 'ys']].values, df[['xe', 'ye']].values)
# create training set
y = np.log(df['len'] * 15 + 1)
# remove non-predictive features
df.drop(['CALL_TYPE', 'TAXI_ID', 'xe', 'ye', 'len'], axis=1, inplace=True)
X = np.array(df, dtype=np.float)
# clean data by removing long distance tracks
th1 = np.percentile(d1, [99.9])
X = X[(d1 < th1), :]
y = y[(d1 < th1)]
print('training a random forest regressor ...')
# Initialize the famous Random Forest Regressor from scikit-learn
clf = RandomForestRegressor(n_estimators=200, n_jobs=-1, random_state=21)
if not os.path.isfile(filename):
continue
df = pd.read_csv(filename)
if df.shape[0] < 1000:
print('skipping key point %i (%i)' % (id_, df.shape[0]))
continue
# factorize categorical columns in training set
#df['CALL_TYPE'], ct_index = pd.factorize(df['CALL_TYPE'])
#df = df[df['CALL_TYPE'] == 0] # A=2, B=1, C=0
# fill all NaN values with -1
#df = df.fillna(-1)
# remove long distance
d1 = haversineKaggle(df[['xs', 'ys']], df[['xe', 'ye']])
th1 = np.percentile(d1, [99.9])
df = df.loc[d1 < th1]
y = np.ravel(np.log(df['len']*15 + 1))
df.drop(['CALL_TYPE', 'TAXI_ID', 'xe', 'ye', 'len'], axis=1, inplace=True)
X = np.array(df, dtype=np.float)
print('training classifier of key point %i (sz=%i) ...' % (id_, X.shape[0]))
# Initialize the famous Random Forest Regressor from scikit-learn
clf = RandomForestRegressor(n_estimators=200, n_jobs=-1, random_state=21)
clf.fit(X, y)
pred_rf = clf.predict(X_tst[id_, :])
clf = GradientBoostingRegressor(n_estimators=200, max_depth=3, random_state=21)
clf.fit(X, y)