ML in Geofisika

import pandas as pd
import numpy as np

# Load well log data
df = pd.read_csv('well_log.csv')

# Check missing values
print(df.isnull().sum())

# Strategy 1: Drop rows with missing values (use with caution)
df_clean = df.dropna()

# Strategy 2: Forward fill (untuk data sequential seperti depth)
df['GR_filled'] = df['GR'].fillna(method='ffill')

# Strategy 3: Interpolation (linear, polynomial, spline)
df['RHOB_interp'] = df['RHOB'].interpolate(method='linear')

# Strategy 4: Fill with median/mean (untuk data non-sequential)
df['NPHI_filled'] = df['NPHI'].fillna(df['NPHI'].median())

# Strategy 5: Predictive imputation using other features
from sklearn.impute import KNNImputer
imputer = KNNImputer(n_neighbors=5)
df_imputed = pd.DataFrame(
    imputer.fit_transform(df[['GR', 'RHOB', 'NPHI']]),
    columns=['GR', 'RHOB', 'NPHI']
)

# Method 1: Statistical approach (Z-score)
from scipy import stats
z_scores = np.abs(stats.zscore(df['GR']))
df_no_outliers = df[z_scores < 3]  # Remove data beyond 3 std

# Method 2: IQR (Interquartile Range)
Q1 = df['RHOB'].quantile(0.25)
Q3 = df['RHOB'].quantile(0.75)
IQR = Q3 - Q1
lower_bound = Q1 - 1.5 * IQR
upper_bound = Q3 + 1.5 * IQR
df_filtered = df[(df['RHOB'] >= lower_bound) & (df['RHOB'] <= upper_bound)]

# Method 3: Domain knowledge (physics-based constraints)
# For RHOB (bulk density), physically impossible values
df_clean = df[(df['RHOB'] > 1.0) & (df['RHOB'] < 3.5)]

# Method 4: Isolation Forest (ML-based outlier detection)
from sklearn.ensemble import IsolationForest
iso_forest = IsolationForest(contamination=0.1, random_state=42)
outlier_labels = iso_forest.fit_predict(df[['GR', 'RHOB', 'NPHI']])
df_clean = df[outlier_labels == 1]  # Keep only inliers

from sklearn.preprocessing import StandardScaler, MinMaxScaler, RobustScaler

# StandardScaler: mean=0, std=1 (sensitive to outliers)
scaler = StandardScaler()
df_scaled = scaler.fit_transform(df[['GR', 'RHOB', 'NPHI']])

# MinMaxScaler: scale to [0,1] range
scaler = MinMaxScaler()
df_normalized = scaler.fit_transform(df[['GR', 'RHOB', 'NPHI']])

# RobustScaler: uses median and IQR (robust to outliers)
scaler = RobustScaler()
df_robust = scaler.fit_transform(df[['GR', 'RHOB', 'NPHI']])

# Log transformation (for skewed data)
df['GR_log'] = np.log1p(df['GR'])  # log1p = log(1+x) to handle zeros

# Box-Cox transformation (requires positive values)
from scipy.stats import boxcox
df['RHOB_boxcox'], lambda_param = boxcox(df['RHOB'])

import lasio

# Read LAS file
las = lasio.read('well_data.las')

# Access well information
print(f"Well Name: {las.well.WELL.value}")
print(f"Field: {las.well.FLD.value}")
print(f"Start Depth: {las.well.STRT.value} {las.well.STRT.unit}")

# Convert to pandas DataFrame
df = las.df()
print(df.head())

# Access specific curves
gr_curve = las.curves['GR']
print(f"GR units: {gr_curve.unit}")
print(f"GR description: {gr_curve.descr}")

# Filter by depth range
df_filtered = df[(df.index >= 2000) & (df.index <= 3000)]

# Export to CSV
df.to_csv('well_log_clean.csv')