File size: 4,559 Bytes
542ac35 | 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 | import os
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.utils.class_weight import compute_class_weight
from src.config import (
DISTORTION_LABELS, LABEL2ID, NUM_LABELS,
CSV_LABEL_MAP, CSV_NO_DISTORTION,
DATA_RAW_DIR, DATA_PROC_DIR, SEED
)
def load_raw(csv_path: str) -> pd.DataFrame:
df = pd.read_csv(csv_path, encoding="utf-8")
rename_map = {}
for col in df.columns:
cl = col.lower().strip()
if "patient question" in cl:
rename_map[col] = "text"
elif "dominant" in cl:
rename_map[col] = "dominant_distortion"
elif "secondary" in cl:
rename_map[col] = "secondary_distortion"
elif "distorted part" in cl:
rename_map[col] = "distorted_part"
df = df.rename(columns=rename_map)
df["text"] = df["text"].astype(str).str.strip()
df["dominant_distortion"] = df["dominant_distortion"].astype(str).str.strip()
if "secondary_distortion" in df.columns:
df["secondary_distortion"] = df["secondary_distortion"].astype(str).str.strip()
else:
df["secondary_distortion"] = ""
return df
def normalise_label(raw: str):
"""Map a CSV label string to our canonical label, or return None."""
raw = str(raw).strip()
if raw == CSV_NO_DISTORTION or raw.lower() == "no distortion":
return None
return CSV_LABEL_MAP.get(raw, raw)
def build_label_vector(dominant: str, secondary: str) -> list:
vec = [0] * NUM_LABELS
dom = normalise_label(dominant)
if dom and dom in LABEL2ID:
vec[LABEL2ID[dom]] = 1
if secondary and secondary not in ("nan", "", "None", "NaN"):
sec = normalise_label(secondary)
if sec and sec in LABEL2ID:
vec[LABEL2ID[sec]] = 1
return vec
def run(csv_filename: str = "Annotated_data.csv") -> None:
csv_path = os.path.join(DATA_RAW_DIR, csv_filename)
os.makedirs(DATA_PROC_DIR, exist_ok=True)
print(f"Loading: {csv_path}")
df = load_raw(csv_path)
print(f"Rows loaded: {len(df)}")
label_vecs = df.apply(
lambda row: build_label_vector(
row["dominant_distortion"], row["secondary_distortion"]
),
axis=1
)
label_df = pd.DataFrame(label_vecs.tolist(), columns=DISTORTION_LABELS)
df = pd.concat(
[df[["text", "dominant_distortion", "secondary_distortion"]], label_df],
axis=1
)
distortion_rows = df[df[DISTORTION_LABELS].sum(axis=1) > 0]
no_distortion_rows = df[df[DISTORTION_LABELS].sum(axis=1) == 0]
multi_label_rows = df[df[DISTORTION_LABELS].sum(axis=1) == 2]
print(f"Rows with >=1 distortion label : {len(distortion_rows)}")
print(f"Rows with no distortion (all 0s) : {len(no_distortion_rows)}")
print(f"Multi-label rows (2 labels) : {len(multi_label_rows)}")
train_df, test_df = train_test_split(
df,
test_size=0.2,
random_state=SEED,
stratify=df["dominant_distortion"]
)
train_df = train_df.reset_index(drop=True)
test_df = test_df.reset_index(drop=True)
print(f"\nTrain size : {len(train_df)}")
print(f"Test size : {len(test_df)}")
train_distortion_only = train_df[train_df[DISTORTION_LABELS].sum(axis=1) > 0]
dom_labels = (
train_distortion_only["dominant_distortion"]
.map(normalise_label)
.dropna()
)
present_labels = np.array(sorted(dom_labels.unique()))
raw_weights = compute_class_weight(
class_weight="balanced",
classes=present_labels,
y=dom_labels.values
)
class_weights = np.ones(NUM_LABELS, dtype=np.float32)
for label, weight in zip(present_labels, raw_weights):
if label in LABEL2ID:
class_weights[LABEL2ID[label]] = weight
print("\nClass weights aligned to DISTORTION_LABELS order:")
for label, w in zip(DISTORTION_LABELS, class_weights):
print(f" {label:<35} {w:.4f}")
train_path = os.path.join(DATA_PROC_DIR, "train.csv")
test_path = os.path.join(DATA_PROC_DIR, "test.csv")
weights_path = os.path.join(DATA_PROC_DIR, "class_weights.npy")
train_df.to_csv(train_path, index=False)
test_df.to_csv(test_path, index=False)
np.save(weights_path, class_weights)
print(f"\nSaved: {train_path}")
print(f"Saved: {test_path}")
print(f"Saved: {weights_path}")
print("\nPreprocessing complete.")
if __name__ == "__main__":
run()
|