/
learned-selection-strategy.py
151 lines (115 loc) · 6.65 KB
/
learned-selection-strategy.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
from __future__ import annotations
import argparse
import os
import time
from collections import defaultdict
import pandas as pd
import numpy as np
from sklearn.ensemble import GradientBoostingRegressor
from util.generator import generate_bw_hist
from util.features import prepare_data
def smape(y_true: np.array, y_pred: np.array) -> np.array:
return 200*np.mean(np.abs(y_pred-y_true)/(np.abs(y_true) + np.abs(y_pred)))
def sape(y_true: np.array, y_pred: np.array) -> np.array:
return 200*(np.abs(y_pred-y_true)/(np.abs(y_true) + np.abs(y_pred)))
def mse(y_true: np.array, y_pred: np.array) -> np.array:
return (y_true - y_pred)**2
def mae(y_true: np.array, y_pred: np.array) -> np.array:
return np.abs(y_true - y_pred)
def rel_err(y_true: np.array, y_pred: np.array) -> np.array:
return (y_pred - y_true).abs() / y_true
def train_model(data_train: pd.DataFrame, data_test: pd.DataFrame, algorithm: str, objective: str, features: list[int], est_range: list[int], d_range: list[int]) -> tuple[GradientBoostingRegressor, pd.DataFrame, pd.DataFrame]:
best_err = np.inf
timings = []
for est in est_range:
for d in d_range:
model = GradientBoostingRegressor(n_estimators = est, max_depth = d, random_state = 1337)
obj_idx = data_train.columns.get_loc(objective)
train_time = time.time()
model.fit(data_train.iloc[:,features].values, data_train.iloc[:,obj_idx])
train_time = time.time() - train_time
pred_time = time.time()
y_pred = model.predict(data_test.iloc[:,features].values)
pred_time = time.time() - pred_time
y_test = data_test.iloc[:,obj_idx]
timings.append((est, d, train_time, pred_time))
eor = pd.DataFrame({"pred": y_pred,
"mse": mse(y_test, y_pred).values,
"mae": mae(y_test, y_pred).values,
"sape": sape(y_test, y_pred).values})
print("""Trained model {}x{} for {} ({}) in {:.4f}s. Predicted in {:.4f}ms ({:.4f}µs per sample)
Errors: MSE={:.4f}, MAE={:.4f}, SMAPE={:.2f}%""". \
format(est, d, algorithm, objective, train_time, pred_time*1000,
pred_time/len(data_test)*1000000, eor["mse"].mean(), eor["mae"].mean(),
eor["sape"].mean()))
if eor["sape"].mean() < best_err:
best_model = model
best_err = eor["sape"].mean()
errors = pd.concat([data_test.iloc[:,features].reset_index(drop=True),
y_test.reset_index(drop=True), eor], axis=1)
timings = pd.DataFrame(timings, columns=['trees', 'depth', 'train time', 'forward pass'])
return best_model, errors, timings
def use_selection_strategy(pool: defaultdict, data_test: pd.DataFrame, objective: str, features: list[int]) -> pd.DataFrame:
data_pred = data_test[["settingIdx", "format", objective]].copy()
for p_algo in data_test["format"].unique():
preds = pool[p_algo][objective].predict(data_test[data_test["format"] == p_algo].iloc[:,features].values)
data_pred.loc[data_pred["format"] == p_algo, ["pred"]] = preds
algo_idx = data_pred.groupby("settingIdx", as_index=False).agg({"pred": "idxmin", objective: "idxmin"})
find_best = pd.concat((data_pred.loc[algo_idx[objective]][["format", objective]].reset_index(drop=True),
data_pred.loc[algo_idx["pred"]][["format", objective]].reset_index(drop=True)),
axis=1)
find_best.columns = ['correct algorithm', 'true', 'predicted algorithm', 'pred']
find_best["equal"] = (find_best["correct algorithm"] == find_best["predicted algorithm"]).astype(int)
find_best["sape"] = sape(find_best["true"], find_best["pred"])
find_best.insert(0, "objective", objective)
print("Objective: {}, accuracy {:.2f}%".format(objective, find_best["equal"].sum()/len(find_best)*100))
print("Objective: {}, slowdown {:.2f}%".format(objective, find_best.loc[find_best["equal"] == 0, "sape"].mean()))
return find_best
def main(argv: list[str] | None = None) -> int:
parser = argparse.ArgumentParser(description='A local learned selection strategy for lightweight integer compression')
parser.add_argument("-g", "--generator", type=str, help='Which generator to be used. Can be laola, outlier, or tidal. Default: laola', default='laola')
parser.add_argument("-d", "--datagen", action='store_true', help='If set, only the data generation is executed.')
args = parser.parse_args(argv)
data_train = {}
data_test = {}
# generate bw histograms in measurements_hist
generate_bw_hist(args.generator, 64, 100)
if args.datagen:
return 0
# discover all algorithms
# measurement data contains the features, objectives and bit width histograms
for f_name in os.listdir("measurements/{}/".format(args.generator)):
data = pd.read_csv("measurements/{}/{}".format(args.generator, f_name), sep="\t")
# feature generation
if "Avg" not in data.columns.values:
data = prepare_data(data, ["bwHist_{}".format(x) for x in range(1,65)])
if "train" in f_name:
print("Found {}.".format(f_name))
data_train[data["format"].iloc[0]] = data
elif "test" in f_name:
data_test[data["format"].iloc[0]] = data
# the model pool
models = defaultdict(dict)
errors = defaultdict(dict)
timings = defaultdict(dict)
obj_columns = ["runtime compr ram2ram [µs]",
"runtime decompr ram2ram [µs]",
"runtime decompr ram2reg [µs]",
"runtime compr cache2ram [µs]",
"compressed size [byte]"]
features = [21,22,23,13,14,15,16,17,18,19]
# train model for each algorithm
for algo in data_train.keys():
assert algo in data_test.keys(), "{} not in test data".format(algo)
# train model for each algorithm x objective
for objective in obj_columns:
models[algo][objective], errors[algo][objective], timings[algo][objective] = train_model(data_train[algo], data_test[algo], algo, objective, features, range(10,60,10), range(3,7))
# test predictions (accuracy and slowdown)
# need full data to find best algorithm for each settingIdx
data_test_full = pd.concat(data_test.values(), ignore_index = True)
for objective in obj_columns:
results = use_selection_strategy(models, data_test_full, objective, features)
results.to_csv("results/{}_{}.csv".format(objective, args.generator), index=False)
return 0
if __name__ == "__main__":
main()