import matplotlib.pyplot as plt
import numpy as np
import scipy.stats
import subprocess

# Names of branches to compare.
BRANCHES = ["base", "alt"]

# Total number of runs for each branch.
RUNS = 250

# Initial runs to discard (50 is probably excessive here)
SKIP = 50

# Get the timings
values = {branch: [] for branch in BRANCHES}
for i, branch in enumerate(BRANCHES * RUNS):
    subprocess.run(["git", "checkout", branch], check=True, capture_output=True)
    subprocess.run(["make"], check=True, capture_output=True)
    res = subprocess.run(f"./python.exe timecomb.py", shell=True, capture_output=True, encoding="utf-8")
    value = float(res.stdout)
    values[branch].append(value)
    print(f"{branch:4s} {i//2}: {value:.3f}ns")

base = values["base"][SKIP:]
alt = values["alt"][SKIP:]

# Plot of the timings
plt.plot(base, label="base")
plt.plot(alt, label="alt")
plt.xlabel("Run index")
plt.ylabel("Time (ns)")
plt.legend()
plt.show()

# Show percent speedup for alt over base.
print(f"Mean time for base: {np.mean(base):.3f}ns")
print(f"Mean for alt: {np.mean(alt):.3f}ns")
speedup_factor = np.mean(base) / np.mean(alt)
print(f"Speedup: {speedup_factor - 1.0:.2%}")

# t-test for the hypothesis that the means are different
print(scipy.stats.ttest_ind(base, alt))