lochan-eda

A practical toolkit for exploring and preparing tabular data for machine learning.

lochan-eda provides two simple workflows:

• Explore a dataset with summaries, plots, missing-value analysis, and a PDF report.
• Prepare tabular data with automatic numerical and categorical preprocessing.

---

Ready to Use

Installation

pip install lochan-eda

1. Automatically Prepare Data for Machine Learning

Use AutomatedEDA when you want to prepare a tabular dataset before training a machine-learning model.

import pandas as pd
from lochan_eda import AutomatedEDA

# Load your data
df = pd.read_csv("data.csv")

eda = AutomatedEDA()

X_train, X_test, y_train, y_test = eda.prepare(
    df,
    target="target"
)

By default, prepare():

• separates the target column
• creates an 80/20 train-test split
• learns preprocessing from the training data
• applies the same learned preprocessing to the test data
• processes numerical and categorical columns separately

You can then pass the processed data directly to your model.

from sklearn.ensemble import RandomForestClassifier

model = RandomForestClassifier(random_state=42)
model.fit(X_train, y_train)

Exclude columns

Keep columns such as IDs out of automatic preprocessing:

eda = AutomatedEDA()

X_train, X_test, y_train, y_test = eda.prepare(
    df,
    target="target",
    exclude=["customer_id"]
)

Use your own target Series

X = df.drop(columns="target")
y = df["target"]

eda = AutomatedEDA()

X_train, X_test, y_train, y_test = eda.prepare(
    X,
    target=y
)

Prepare without a train-test split

X, y = eda.prepare(
    df,
    target="target",
    split=False
)

If there is no target, prepare() can also process feature data only:

X = eda.prepare(
    df,
    split=False
)

---

2. Profile a Dataset

Use Profiler when you want to understand a dataset before modeling.

import pandas as pd
from lochan_eda import Profiler

df = pd.read_csv("data.csv")

profile = Profiler(df, target="target")

Dataset overview

profile.overview()

This returns information about:

• rows and columns
• memory usage
• duplicate rows
• duplicate percentage
• missing cells
• missing percentage
• numerical columns
• categorical columns
• datetime columns

Numerical analysis

profile.numerical.summary()
profile.numerical.plot()

Categorical analysis

profile.categorical.summary()
profile.categorical.plot()

Missing-value analysis

profile.missing.plot()

Generate a PDF report

profile.report.save("eda_report.pdf")

The report contains dataset overview information, numerical and categorical summaries, and generated analysis plots.

---

3. Analyze Only Selected Columns

Numerical and categorical plots can be limited to selected columns.

profile.numerical.plot(
    columns=["age", "income"]
)

profile.categorical.plot(
    columns=["city", "education"],
    top_n=10
)

top_n keeps the most frequent categories visible and groups the remaining categories as Others.

---

API Overview

AutomatedEDA

from lochan_eda import AutomatedEDA

The main interface for automatic tabular-data preprocessing.

AutomatedEDA()

AutomatedEDA()

Creates an automatic preprocessing object.

prepare()

prepare(
    X,
    target=None,
    exclude=None,
    split=True,
    test_size=0.2,
    random_state=42,
    stratify=None
)

The high-level method for preparing data.

Parameter	Description
X	Input pandas.DataFrame.
target	Target column name or pandas.Series.
exclude	Column name or list of columns to exclude from preprocessing.
split	Whether to create train/test data. Default: True.
test_size	Proportion used for the test set. Default: 0.2.
random_state	Random state for reproducibility. Default: 42.
stratify	Values used for stratified splitting.

Return values

The returned values depend on target and split:

target	split	Returns
Not provided	False	X
Not provided	True	X_train, X_test
Provided	False	X, y
Provided	True	X_train, X_test, y_train, y_test

fit()

fit(X, y=None)

Learns preprocessing rules from the supplied data.

fit() separates numerical and categorical columns and learns the required transformations for each type.

transform()

transform(X, y=None)

Applies preprocessing rules learned by fit().

transform() must be called after fit().

---

Profiler

from lochan_eda import Profiler

Provides dataset-level exploratory analysis.

Profiler()

Profiler(df, target=None)

Parameter	Description
df	Input pandas.DataFrame.
target	Target column name.

When target is a column name, the target is separated from the profiling data.

overview()

profile.overview()

Prints and returns a dictionary containing:

• rows
• columns
• memory_usage
• duplicate_rows
• duplicate_percentage
• missing_cells
• missing_percentage
• numerical_columns
• categorical_columns
• datetime_columns

---

Numerical

from lochan_eda import Numerical

Handles numerical-column profiling and preprocessing.

Numerical()

Numerical(profiler_df=None)

profiler_df is used by the profiling methods such as summary() and plot().

fit()

fit(data, exclude=None)

Learns numerical preprocessing rules from the supplied data.

The numerical workflow can learn rules for:

• missing-value imputation
• column removal based on missingness
• outlier handling
• scaling

transform()

transform(data, exclude=None)

Applies the rules learned by fit().

summary()

profile.numerical.summary()

Returns a DataFrame containing descriptive numerical statistics together with missing-value, skewness, zero, and outlier information.

plot()

profile.numerical.plot(columns=None)

Creates numerical analysis plots for each selected numerical column:

• distribution
• box plot
• Q-Q plot

The figure is also saved as numerical_plots.png.

imputer()

imputer(learn=False)

Internal numerical preprocessing step for learning or applying missing-value handling.

outlier_manager()

outlier_manager(learn=False)

Internal numerical preprocessing step for learning or applying outlier rules.

scaler()

scaler(learn=False)

Internal numerical preprocessing step for learning or applying feature scaling.

imputer(), outlier_manager(), and scaler() are lower-level methods. For normal usage, prefer AutomatedEDA or Numerical.fit() / Numerical.transform().

---

Categorical

from lochan_eda import Categorical

Handles categorical-column profiling and preprocessing.

Categorical()

Categorical(profiler_df=None)

profiler_df is used by the profiling methods.

fit()

fit(data, exclude=None)

Learns categorical preprocessing rules.

The categorical workflow can learn rules for:

• missing-value handling
• rare-category handling
• binary encoding
• one-hot encoding
• frequency encoding

transform()

transform(data, exclude=None)

Applies the categorical preprocessing rules learned by fit().

summary()

profile.categorical.summary()

Returns descriptive statistics for categorical columns.

plot()

profile.categorical.plot(columns=None, top_n=10)

Creates category-distribution bar charts.

The figure is also saved as categorical_plots.png.

imputer()

imputer(learn=False)

Internal categorical preprocessing step for learning or applying missing-value handling.

rare_manager()

rare_manager(learn=False)

Internal preprocessing step that learns or applies rare-category grouping.

encoder()

encoder(learn=False)

Internal preprocessing step that selects and applies categorical encoding based on category cardinality.

imputer(), rare_manager(), and encoder() are lower-level methods. For normal usage, prefer AutomatedEDA or Categorical.fit() / Categorical.transform().

---

Missing

from lochan_eda.missing import Missing

Provides missing-value visualization.

Missing()

Missing(profiler_df=None)

Creates a missing-value analysis object.

plot()

profile.missing.plot()

Creates a horizontal bar chart showing missing-value percentages by column.

The figure is also saved as missing_plot.png.

---

Report

from lochan_eda.report import Report

Generates an EDA PDF report from a Profiler instance.

Report()

Report(profiler)

save()

profile.report.save("report.pdf")

Generates and saves the exploratory data analysis report.

Default path:

profile.report.save()

which creates:

report.pdf

---

Utility Functions

The package also contains helper functions in lochan_eda.utils.

get_iqr_bounds()

from lochan_eda.utils import get_iqr_bounds

lower, upper = get_iqr_bounds(series)

Returns lower and upper bounds calculated using the IQR method.

get_active_cols()

from lochan_eda.utils import get_active_cols

columns = get_active_cols(all_cols, exclude=["id"])

Returns columns after removing excluded columns.

is_matching()

from lochan_eda.utils import is_matching

is_matching(fit_df, transform_df)

Checks whether the two DataFrames contain matching column sets.

---

Typical Workflow

A common machine-learning workflow with lochan-eda looks like this:

import pandas as pd
from lochan_eda import AutomatedEDA
from sklearn.ensemble import RandomForestClassifier

# 1. Load data
df = pd.read_csv("data.csv")

# 2. Prepare data
eda = AutomatedEDA()
X_train, X_test, y_train, y_test = eda.prepare(
    df,
    target="target"
)

# 3. Train your model
model = RandomForestClassifier(random_state=42)
model.fit(X_train, y_train)

# 4. Predict
predictions = model.predict(X_test)

For exploration before modeling:

from lochan_eda import Profiler

profile = Profiler(df, target="target")

profile.overview()
profile.numerical.summary()
profile.categorical.summary()
profile.missing.plot()
profile.report.save("eda_report.pdf")

---

Package

• Package: lochan-eda
• Author: Lochan Jangid
• Version: 0.2.0