#! /usr/bin/env python
from format_data import formatData
from train_xgboost import train
import csv
def loadFinalData(flag):
dataCSV = csv.reader(open("../data/data_"+flag+".csv", "r"))
data = [row for row in dataCSV]
formatData(data[1:], flag)
loadFinalData("train")
loadFinalData("test")
train("final")
from format_data import formatData
from train_xgboost import train
import csv
dataCSV = csv.reader(open("../data/data_train.csv", "r"))
data = [row for row in dataCSV]
del data[0]
m = (len(data)-1)//10+1
data_origin = []
for i in range(len(data)):
data_origin.append(data[i][:2])
for k in range(10):
left, right = k*m, min((k+1)*m, len(data))
realFile = open("real_value_"+str(k)+".csv", "w")
lines = ["LKADT_P,DEATH\n"]
for i in range(left, right):
lines.append("{},{}\n".format(data[i][0], data[i][1]))
realFile.writelines(lines)
formatData(data[:left]+data[right:], "train")
for i in range(left, right):
data[i][0], data[i][1] = 0, 1
formatData(data[left:right], "test")
for i in range(left, right):
data[i][0], data[i][1] = data_origin[i]
train(str(k))
value_train_dir = "{}/stage*/train_value".format(args.outdir)
policy_train_dir = "{}/stage*/train_policy".format(args.outdir)
value_train_dirs = glob.glob(value_train_dir)
policy_train_dirs = glob.glob(policy_train_dir)
T0 = time.time()
if args.model_type == "xgboost":
value_train_dirs = [d + "/svmlight" for d in value_train_dirs]
policy_train_dirs = [d + "/svmlight" for d in policy_train_dirs]
from train_xgboost import train
if args.target_model in ("value", "all"):
value_modelfile = "{}/value_xgb".format(args.outdir)
print("\n\nTraining value model")
train(args,
value_train_dirs,
value_modelfile,
n_features,
objective="reg:squarederror",
oversample=True)
if args.target_model in ("policy", "all"):
policy_modelfile = "{}/policy_xgb".format(args.outdir)
print("\n\nTraining policy model")
train(args,
policy_train_dirs,
policy_modelfile,
n_features,
objective="reg:squarederror",
oversample=True)
elif args.model_type == "nn":
value_train_dirs = [d + "/svmlight" for d in value_train_dirs]
policy_train_dirs = [d + "/svmlight" for d in policy_train_dirs]
def train_xgboost(opt):
import train_xgboost as xgboost
team_data = get_team_representations(opt.team_data_type)
xgboost.train(team_data, opt)