import pandas as pd

import matplotlib.pyplot as plt

# --- CONFIG -------------------------------------------------------------

CSV_PATH = "new_results_with_seed.csv" # ← change to your file name if different

OUT_PNG = "graficos/Acurácia por Ruido.png"

# ------------------------------------------------------------------------

# 1) Load the CSV

df = pd.read_csv(CSV_PATH)

# 2) Filter relevant rows

# Fixando a Label, o Round e o Método para consistência

mask = (df["Label"] == 0) & \

(df["Round"] == 50) & \

(df["Method"] == "loss") & \

(df["Noise Scale"] < 0.11)

subset = df.loc[mask, ["Noise Scale", "Accuracy"]].copy()

# 3) Convert to numeric

subset["Noise Scale"] = pd.to_numeric(subset["Noise Scale"], errors="coerce")

subset["Accuracy"] = pd.to_numeric(subset["Accuracy"], errors="coerce")

# 4) Normalize accuracy to [0, 1]

subset["Accuracy"] /= 100

# 5) Group by Noise Scale

grouped = subset.groupby("Noise Scale")["Accuracy"]

means = grouped.mean()

stds = grouped.std()

counts = grouped.count()

print(counts)

# Corrigir os desvios padrão para erro padrão da média (±1.96 * std / sqrt(n))

stds = 1.96 * stds / counts.pow(0.5)

# 6) Plot with error bars

plt.figure(figsize=(8, 5))

plt.errorbar(

means.index, means.values, yerr=stds.values,

fmt='o-', capsize=5

)

# 7) Labels and formatting

plt.xlabel("Ruído", fontsize=16)

plt.ylabel("Acurácia média", fontsize=16)

plt.ylim(0.5, 1)

plt.grid(True)

plt.tight_layout()

plt.savefig(OUT_PNG, dpi=300)

plt.close()