INTERFACES / CLI DEEP DIVE & REFERENCE

The Command Line Interface, Completely.

DATADOC provides a local-first, zero-leakage CLI for tabular dataset preparation. Explore parameters, execution mechanics, command breakdowns, and interactive demo testing.

datadoc --help Polars Powered Zero Data Leakage v0.6.1

📥 Download demo.csv Explore Transformation Sandbox ↓

0. Interactive Demo Sandbox & Test Drive

Download our official 500-row demo.csv test dataset and watch how DATADOC profiles, plans, fits, transforms, and evaluates the dataset step-by-step.

Download Test Dataset (demo.csv)

Includes missing values, scale mismatches, UUID identifiers, string dates, and extreme outliers (age=999).

📥 Download demo.csv (46 KB)

Raw Input Dataset: `demo.csv` (First 5 Rows)

Notice the dirty features requiring engineering before machine learning model training:

customer_id (ID)	name (Text)	age	salary	department	city	signup_date	churn (Target)
`dba51b99...`	Person_0	23	57,098	Engineering	Hyderabad	2020-07-10	0
`4f57a639...`	Person_1	56	83,878	Engineering	Bangalore	2021-11-20	0
`10eb978b...`	Person_2	60	30,695	Support	Chennai	2023-07-11	1
`e22b64ba...`	Person_3	null	62,930	Sales	Chennai	2022-01-28	1
`b88ae03d...`	Person_4	999 (Outlier)	47,431	Sales	Bangalore	2023-02-14	0

Step 1: `datadoc profile demo.csv --target churn`

Profiles dataset health, missing value counts, cardinality, column roles, and generates a unique schema fingerprint.

JSONdatadoc profile output
{
  "rows": 500,
  "columns": 10,
  "schema": {
    "customer_id": "String",
    "name": "String",
    "age": "Int64",
    "salary": "Int64",
    "department": "String",
    "city": "String",
    "signup_date": "String",
    "churn": "Int64"
  },
  "null_counts": { "age": 25, "salary": 20, "department": 15 },
  "roles": [
    { "name": "customer_id", "role": "identifier" },
    { "name": "name", "role": "identifier" },
    { "name": "age", "role": "feature_numeric" },
    { "name": "salary", "role": "feature_numeric" },
    { "name": "department", "role": "feature_categorical" },
    { "name": "churn", "role": "target" }
  ],
  "schema_fingerprint": "a3f890e12d..."
}

Step 2: `datadoc plan demo.csv --target churn --drop-identifiers`

Formulates an explicit, explainable sequence of transformation operations with analytical rationales before modifying data.

JSONdatadoc plan output
{
  "target": "churn",
  "operations": [
    {
      "name": "DropIdentifiers",
      "columns": ["customer_id", "name"],
      "rationale": "High-cardinality string IDs overfit models and carry zero predictive signal."
    },
    {
      "name": "ImputeNumeric",
      "columns": ["age", "salary"],
      "strategy": "median",
      "rationale": "Fill missing numerical values with training median (outlier-resistant)."
    },
    {
      "name": "ImputeCategorical",
      "columns": ["department"],
      "strategy": "mode",
      "rationale": "Fill missing categorical entries with the most frequent value."
    },
    {
      "name": "DatetimeExtract",
      "columns": ["signup_date"],
      "features": ["month", "day", "dayofweek"],
      "rationale": "Extract temporal sub-components from date strings."
    },
    {
      "name": "OneHotEncode",
      "columns": ["department", "city"],
      "rationale": "Convert categorical strings into numeric binary indicator vectors."
    }
  ]
}

Step 3: `datadoc fit demo.csv --target churn --scaling standard --clip-outliers --output pipeline.json`

Learns medians, modes, Z-score parameters, and IQR bounds strictly from training data and saves pipeline.json.

JSONpipeline.json artifact
{
  "artifact_version": 1,
  "config": {
    "target": "churn",
    "drop_identifiers": true,
    "scaling": "standard",
    "clip_outliers": true
  },
  "state": {
    "dropped": ["customer_id", "name"],
    "medians": { "age": 36.0, "salary": 78500.0 },
    "modes": { "department": "Engineering" },
    "scaling": {
      "age": { "mean": 36.4, "std": 11.2 },
      "salary": { "mean": 79200.0, "std": 31500.0 }
    },
    "outliers": {
      "age": { "lower": 12.0, "upper": 75.0 }
    }
  }
}

Step 4: `datadoc transform demo.csv --pipeline pipeline.json --output clean_demo.csv`

Before vs After summary showing how dirty features transformed into clean, model-ready numerical columns:

Feature Column	Raw Input State	Transformed Output State	Transformation Applied
`customer_id` / `name`	UUIDs & Names	Dropped	Identified as string IDs and removed
`age`	Nulls + Outlier (999)	`-1.19` (Z-score)	Null → Median (36.0), Clipped (75.0), Standardized
`salary`	Nulls present	`-0.68` (Z-score)	Null → Median (78,500), Standardized
`department`	"Engineering"	`dept_Engineering=1`, `dept_Sales=0`	One-Hot Encoded into binary indicator vectors
`signup_date`	"2020-07-10"	`month=7`, `day=10`, `dayofweek=4`	Extracted numeric temporal features
`churn` (Target)	`0` or `1`	`0` or `1` (Untouched)	Protected target column preserved exactly

Step 5: `datadoc evaluate demo.csv --target churn --task classification --estimator tree`

Runs 80/20 train/test evaluation benchmark to measure clean vs raw accuracy improvement:

JSON & Terminaldatadoc evaluate benchmark
{
  "task": "classification",
  "metric": "balanced_accuracy",
  "baseline_score": 0.542,
  "selected_score": 0.815,
  "improvement": +0.273,
  "selected_pipeline": "candidate",
  "feature_count": 16
}

Baseline Score : ▇▇▇▇▇▇▇▇▇▇▇▇▇ 0.542
Candidate Score: ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 0.815 (+27.3% boost!)

1. Core Concepts & Terminology

Before running CLI commands, it is essential to understand the key parameters and arguments that govern DATADOC's behavior.

Target Column (`--target`)

The target is the column you intend to predict (e.g., churn, salary, price). When you declare --target <column>, DATADOC applies a protection boundary to that column. The target is never imputed, scaled, encoded, or clipped. It passes through unchanged so your evaluation metrics and model training remain uncorrupted.

Example: In a subscription dataset, churn contains 0 (customer stayed) or 1 (customer left). Passing --target churn prevents DATADOC from scaling or modifying these class labels.

Task Type (`--task`)

Defines the nature of the supervised learning task. DATADOC uses this to determine evaluation metrics and optimal default preprocessing routines:

auto (Default): Automatically infers task type. If the target has ≤ 20 unique values or is boolean/string, it selects classification; otherwise, regression.
classification: Supervised binary or multi-class prediction (evaluates via balanced accuracy / log loss).
regression: Continuous numeric prediction (evaluates via RMSE / R²).

Identifier Dropping (`--drop-identifiers`)

High-cardinality string IDs (such as customer_id, uuid, user_name) carry no generalizable predictive signal and cause severe overfitting if one-hot encoded. Passing --drop-identifiers instructs DATADOC to detect string IDs and remove them during transformation.

Feature Scaling Strategies (`--scaling`)

Tabular numeric features often vary by orders of magnitude (e.g., age spans 18–80 while income spans 20,000–500,000). Scaling normalizes these ranges:

Mode	Mathematical Formula	Best Used For
`auto`	Picks `standard` for linear models; `none` for tree models	Default — recommended for most workflows
`none`	`X_out = X`	Tree-based algorithms (RandomForest, XGBoost, LightGBM)
`standard`	`Z = (X - μ) / σ`	Linear Regression, Logistic Regression, Neural Networks, SVMs
`robust`	`Z = (X - Q₂) / IQR`	Datasets containing unclipped, heavy-tailed outliers

Outlier Clipping (`--clip-outliers`)

When enabled, DATADOC calculates upper and lower Interquartile Range (IQR) bounds during fit and clips extreme values during transform:

Lower Bound = Q₁ - (1.5 × IQR)
Upper Bound = Q₃ + (1.5 × IQR)

Values below the lower bound are clamped to the lower bound; values above the upper bound are clamped to the upper bound.

Split Strategies (`--time-column` & `--group-column`)

--time-column <col>: Performs an ordered temporal split (oldest rows in train, newest in test) to prevent future-data leakage in time-series tabular problems.
--group-column <col>: Performs group-aware splitting so that all records belonging to a single group (e.g., all visits by patient_id) reside strictly in train or test.

2. The Pipeline Lifecycle

DATADOC operates on a strict 4-stage lifecycle designed to guarantee zero data leakage between training, validation, and production inference sets.

flowchart LR A["Raw Dataset (CSV/Parquet)"] --> B["datadoc profile"] B --> C["datadoc plan"] C --> D["datadoc fit (Train Data Only)"] D --> E["pipeline.json (Artifact)"] E --> F["datadoc transform"] F --> G["Clean Dataset"]

01 Profile Inspect schema, missingness, cardinality, data roles, and compute a schema fingerprint.

02 Plan Formulate a deterministic sequence of operations with explicit analytical rationales.

03 Fit Learn parameters (medians, mode vocabularies, scaling means/stds) strictly from training rows and persist to JSON.

04 Transform Apply the saved JSON artifact immutably to validation, test, or production inference data.

3. Read-Only Commands

PROFILE datadoc profile <file_path>

Inspects a CSV or Parquet file without making any modifications. Generates detailed health reports, column role assignments, missing value counts, and schema fingerprints.

TerminalInspect dataset quality
datadoc profile demo.csv --target churn --output profile.json

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Path to CSV or Parquet input file
`--target`	Option	`None`	Declare prediction target column to apply protection
`--output`	Option	`None`	Path to write profile output JSON

PLAN datadoc plan <file_path>

Generates a transparent, step-by-step transformation plan without modifying the dataset or learning statistics.

TerminalGenerate transformation plan
datadoc plan demo.csv --target churn --drop-identifiers --output plan.json

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Input CSV or Parquet dataset
`--target`	Option	`None`	Protected target column name
`--task`	Option	`auto`	`auto`, `classification`, or `regression`
`--drop-identifiers`	Flag	`False`	Include identifier dropping in transformation operations
`--output`	Option	`None`	Path to save plan JSON file

REPORT datadoc report <file_path>

Generates an automated, standalone, interactive HTML data health and preparation audit report for sharing with teammates. Completely self-contained with health grade scoring, distribution histograms, null rates, and transformation plans.

TerminalGenerate standalone HTML report
datadoc report train.csv --target churn --output report.html

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Path to CSV or Parquet input dataset
`--target`	Option	`None`	Declare target column for protection and analysis
`--output`	Option	`<stem>_report.html`	Path to write standalone HTML report
`--title`	Option	`None`	Custom title for report header
`--preset`	Option	`None`	Preset to base recommendations on
`--ai`	Flag	`False`	Include AI Executive Summary & Feature Hypotheses in report
`--ai-model`	Option	`None`	Model to use for AI report analysis
`--open / --no-open`	Flag	`True`	Open in browser automatically

COMPARE datadoc compare <raw_path> <transformed_path>

Visually compares raw vs. transformed datasets side-by-side. Displays dimension changes, missing value reduction (100% resolution tracking), column lifecycle (retained, dropped, engineered), and numeric distribution shifts.

TerminalCompare datasets side-by-side
datadoc compare raw.csv clean.csv --target churn --html comparison.html

Options

Option	Type	Default	Description
`raw_path`	Argument	Required	Path to raw CSV or Parquet file
`transformed_path`	Argument	Required	Path to transformed CSV or Parquet file
`--target`	Option	`None`	Target column name
`--html`	Option	`None`	Path to save visual HTML comparison page
`--json`	Flag	`False`	Output comparison data as structured JSON
`--open`	Flag	`False`	Open HTML comparison in browser

EXPLAIN datadoc explain <file_path>

AI-powered dataset feature engineering hypotheses, semantic role inference, missingness mechanism analysis (MCAR/MAR/MNAR), and target leakage audit using LiteLLM (OpenAI, Gemini, Anthropic, Ollama).

TerminalAI Feature Hypotheses & Audit
datadoc explain raw.csv --target churn --model gpt-4o-mini --recommend-config

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Path to CSV or Parquet file to analyze
`--target`	Option	`None`	Target column for predictive task
`--model`	Option	`None`	LLM model (e.g., `gpt-4o-mini`, `gemini/gemini-2.0-flash`, `ollama/llama3`)
`--output`	Option	`None`	Save AI markdown explanation to file
`--recommend-config`	Flag	`False`	Generate and save recommended `datadoc.toml`

HEALTH datadoc health <file_path>

Fast dataset health audit displaying 0–100 quality score, letter grade (A+ to F), and detected data quality findings.

TerminalQuick health check
datadoc health demo.csv --target churn

4. Build Commands

FIT datadoc fit <file_path>

Fits transformation rules (imputation medians, encodings, clipping bounds) strictly on the provided training split and saves a serializable pipeline.json artifact.

TerminalFit pipeline on training data
datadoc fit train.csv \
  --target churn \
  --task classification \
  --drop-identifiers \
  --scaling standard \
  --clip-outliers \
  --output artifacts/pipeline.json

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Path to training CSV/Parquet file
`--target`	Option	`None`	Supervised target column
`--task`	Option	`auto`	`auto`, `classification`, `regression`
`--drop-identifiers`	Flag	`False`	Drop string IDs during pipeline execution
`--scaling`	Option	`auto`	`auto`, `none`, `standard`, `robust`
`--clip-outliers`	Flag	`False`	Opt into upper/lower IQR outlier clipping
`--output`	Option	`pipeline.json`	Path to save fitted JSON artifact

TRANSFORM datadoc transform <file_path>

Applies a saved, fitted pipeline artifact to validation, test, or production inference data. Guarantees that validation rows are transformed using training-derived statistics.

TerminalApply pipeline to validation data
datadoc transform test.csv \
  --pipeline artifacts/pipeline.json \
  --output clean_test.parquet

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Path to dataset to transform
`--pipeline`	Option	Required	Path to fitted `pipeline.json` artifact
`--output`	Option	Required	Output file path (`.csv` or `.parquet`)

EXPORT datadoc export

Exports a standalone, self-contained Python script (pipeline.py) wrapping the fitted artifact for zero-dependency inference in production environments.

TerminalExport standalone Python script
datadoc export \
  --pipeline artifacts/pipeline.json \
  --output my_pipeline.py

Options

Option	Type	Default	Description
`--pipeline`	Option	Required	Path to fitted `pipeline.json` artifact
`--output`	Option	`pipeline.py`	Path to write standalone Python script

5. Measure & Orchestration Commands

EVALUATE datadoc evaluate <file_path>

Performs a strict 80/20 train/test evaluation comparing a baseline model on raw data against a candidate model trained on DATADOC-cleaned data.

flowchart TD Data["Raw Dataset"] --> Split["80 / 20 Train-Test Split"] Split --> RawTrain["80% Raw Train"] Split --> RawTest["20% Raw Test"] RawTrain --> BaseFit["Train Baseline Model"] RawTest --> BaseEval["Evaluate Baseline Score"] RawTrain --> PipeFit["Fit Pipeline (Leakage-Safe)"] PipeFit --> CleanTrain["Transform 80% Train"] PipeFit --> CleanTest["Transform 20% Test"] CleanTrain --> CandFit["Train Candidate Model"] CleanTest --> CandEval["Evaluate Candidate Score"] BaseEval --> Comp["Compare & Report Delta"] CandEval --> Comp

TerminalRun evaluation benchmark
datadoc evaluate demo.csv \
  --target churn \
  --task classification \
  --estimator tree \
  --output eval_report.json

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Input CSV or Parquet file
`--target`	Option	Required	Supervised target column to evaluate
`--task`	Option	`auto`	`auto`, `classification`, `regression`
`--estimator`	Option	`linear`	`linear`, `tree`, or `both`
`--time-column`	Option	`None`	Column name for ordered temporal splits
`--group-column`	Option	`None`	Column name for group-aware splits
`--output`	Option	`None`	Path to save evaluation report JSON

RUN datadoc run <file_path>

Orchestrates a complete, one-shot local run: profiles, plans, fits, transforms, and generates a reproducible run directory with manifest files.

TerminalExecute reproducible local run
datadoc run demo.csv \
  --target churn \
  --output-dir datadoc-run \
  --evaluate

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Input dataset path
`--target`	Option	`None`	Target column name
`--task`	Option	`auto`	`auto`, `classification`, `regression`
`--output-dir`	Option	`datadoc-run`	Directory path to persist run artifacts
`--evaluate`	Flag	`False`	Include evaluation benchmark in run output

6. Dashboard & Utility Commands

UI datadoc ui <file_path>

Launches a local FastAPI backend server and opens the interactive Web Dashboard in your default browser.

TerminalLaunch Web Dashboard
datadoc ui demo.csv --port 8000

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Dataset file to load in dashboard
`--port`	Option	`8000`	HTTP port for local dashboard server

VERSION datadoc version

Displays the current installed version of datadoc-cli.

TerminalCheck DATADOC version
datadoc version

WIZARD datadoc wizard <file_path>

Guided setup: asks for target, preset, identifier/dedup policy, scaling, clipping, and rare-category frequency, writes datadoc.toml, then runs the full pipeline.

TerminalGuided first run
datadoc wizard train.csv --output-dir datadoc-run

Options

Option	Type	Default	Description
`file_path`	Argument	Required	Training CSV or Parquet file
`--output-dir`	Option	`datadoc-run`	Directory for run outputs

INIT datadoc init

Writes a starter datadoc.toml so runs are reproducible without retyping flags. Settings can also live under [tool.datadoc] in pyproject.toml; explicit flags always win.

TerminalRepeatable config
datadoc init --preset balanced --output datadoc.toml

LINT datadoc lint <file_path>

Lints a dataset for leakage and prep pitfalls: target duplication, null targets, infinite values, duplicate rows, constants, and profile findings.

TerminalLeakage lint
datadoc lint train.csv --target churn

DIFF datadoc diff <file_a> <file_b>

Diffs two profile, plan, or pipeline JSON artifacts top-level key by key, plus added/removed plan operations.

TerminalCompare artifacts
datadoc diff plan-v1.json plan-v2.json

PLUGINS datadoc plugins list

Lists the 7 registered deterministic plugins in priority order, including third-party entry-point plugins. Use datadoc plugins show <Name> for detail.

TerminalRegistry introspection
datadoc plugins list

7. Step-by-Step Hands-On Tutorial

Follow this sequence using the downloadable demo.csv file to test all features of the DATADOC CLI. Prefer guidance? Start with datadoc wizard demo.csv instead of steps 1–3.

BashFull CLI workflow test sequence
# Step 1: Inspect dataset quality and health findings
datadoc profile demo.csv --target churn

# Step 2: Preview the proposed transformation plan
datadoc plan demo.csv --target churn --drop-identifiers

# Step 3: Fit the pipeline on training data and save artifact
datadoc fit demo.csv \
  --target churn \
  --drop-identifiers \
  --scaling standard \
  --clip-outliers \
  --output artifacts/pipeline.json

# Step 4: Transform new validation dataset using saved artifact
datadoc transform demo.csv \
  --pipeline artifacts/pipeline.json \
  --output clean_demo.csv

# Step 5: Benchmark candidate vs baseline ML performance (with ablation)
datadoc evaluate demo.csv \
  --target churn \
  --task classification \
  --estimator tree \
  --ablation

# Step 6: Export executable Python wrapper script
datadoc export \
  --pipeline artifacts/pipeline.json \
  --output my_pipeline.py

# Step 7: Run complete reproducible local run in one command
datadoc run demo.csv \
  --target churn \
  --output-dir full-run \
  --evaluate

# Step 8: Open interactive Web Dashboard
datadoc ui demo.csv

8. Master Options Reference Table

Below is a quick reference matrix of all options across DATADOC CLI commands:

Flag / Option	Commands Supported	Allowed Values	Default	Description
`--target`	profile, plan, fit, evaluate, run	Column Name	`None`	Declares protected prediction target column
`--task`	plan, fit, evaluate, run	`auto`, `classification`, `regression`	`auto`	Specifies machine learning problem type
`--drop-identifiers`	plan, fit, run	Boolean Flag	`False`	Detects and removes high-cardinality string IDs
`--identifier-column`	plan, fit, run	Column Name (repeatable)	`None`	Forces a column into the identifier role (e.g. `PassengerId`)
`--ignore-column`	plan, fit, run	Column Name (repeatable)	`None`	Excludes a column from features entirely
`--scaling`	fit, run	`auto`, `none`, `standard`, `robust`	`auto`	Selects numeric feature scaling strategy
`--clip-outliers`	fit, run	Boolean Flag	`False`	Clamps numerical outliers to IQR bounds
`--estimator`	evaluate	`linear`, `tree`, `both`	`linear`	Selects model family; `both` fits linear and tree and keeps the stronger validation score
`--preset`	profile, plan, fit, evaluate, run	`quick`, `balanced`, `linear`, `tree`, `time`, `robust`	`None`	One-word policy bundle for scaling, clipping, threshold, and estimator family
`--deduplicate`	fit, run	Boolean Flag	`False`	Drops duplicate rows from training data at fit time
`--rare-frequency`	fit, run	Float 0–1	`0.0`	Groups categories below this frequency into `__RARE__`
`--cyclical`	fit, run	Boolean Flag	`False`	Adds sin/cos pairs for datetime parts
`--no-hour`	fit, run	Boolean Flag	`False`	Skips hour extraction even when time data is present
`--categorical-threshold`	fit, run	Integer	`20`	Max one-hot width before falling back to frequency encoding
`--ablation`	evaluate, run	Boolean Flag	`False`	Compares full, no-clip, no-scaling, and minimal variants
`--validate`	transform	Boolean Flag	`False`	Runs schema validation and median-shift drift checks
`--explain`	profile, plan	Boolean Flag	`False`	Renders human-readable tables instead of raw JSON
`--compare`	profile	File Path (`.json`)	`None`	Compares against another profile JSON
`--diff`	plan	File Path (`.json`)	`None`	Diffs against another plan JSON
`--config`	profile, plan, fit, evaluate, run	File Path (`.toml`)	Auto-discovered	Explicit `datadoc.toml` path; explicit flags override it
`--format`	export	`python`, `joblib`	`python`	Executable wrapper script or portable joblib artifact dict
`--time-column`	evaluate	Column Name	`None`	Enforces ordered temporal train/test split
`--group-column`	evaluate	Column Name	`None`	Enforces group-aware train/test split
`--pipeline`	transform, export	File Path (`.json`)	Required	Path to fitted pipeline JSON artifact
`--output`	profile, plan, fit, transform, evaluate, export	File Path	Varies	Path to write command output artifact
`--output-dir`	run	Directory Path	`datadoc-run`	Directory path for reproducible run outputs
`--port`	ui	Integer	`8000`	HTTP port for local dashboard server (`--no-browser` skips auto-open)

9. Error Recovery & Best Practices

Holdout Score Warning in Evaluation

Evaluation scores are computed via an 80/20 holdout split. Use an external final test set for critical production decisions.

Missing Input Columns during Transform

If datadoc transform fails with a schema mismatch error, ensure the input dataset contains all feature columns present when datadoc fit was executed.

The Command Line Interface, Completely.

0. Interactive Demo Sandbox & Test Drive

Download Test Dataset (demo.csv)

Raw Input Dataset: demo.csv (First 5 Rows)

Step 1: datadoc profile demo.csv --target churn

Step 2: datadoc plan demo.csv --target churn --drop-identifiers

Step 3: datadoc fit demo.csv --target churn --scaling standard --clip-outliers --output pipeline.json

Step 4: datadoc transform demo.csv --pipeline pipeline.json --output clean_demo.csv

Step 5: datadoc evaluate demo.csv --target churn --task classification --estimator tree

1. Core Concepts & Terminology

Target Column (--target)

Task Type (--task)

Identifier Dropping (--drop-identifiers)

Feature Scaling Strategies (--scaling)

Outlier Clipping (--clip-outliers)

Split Strategies (--time-column & --group-column)

2. The Pipeline Lifecycle

3. Read-Only Commands

Options

Options

Options

Options

Options

4. Build Commands

Options

Options

Options

5. Measure & Orchestration Commands

Options

Options

6. Dashboard & Utility Commands

Options

Options

7. Step-by-Step Hands-On Tutorial

8. Master Options Reference Table

9. Error Recovery & Best Practices

Raw Input Dataset: `demo.csv` (First 5 Rows)

Step 1: `datadoc profile demo.csv --target churn`

Step 2: `datadoc plan demo.csv --target churn --drop-identifiers`

Step 3: `datadoc fit demo.csv --target churn --scaling standard --clip-outliers --output pipeline.json`

Step 4: `datadoc transform demo.csv --pipeline pipeline.json --output clean_demo.csv`

Step 5: `datadoc evaluate demo.csv --target churn --task classification --estimator tree`

Target Column (`--target`)

Task Type (`--task`)

Identifier Dropping (`--drop-identifiers`)

Feature Scaling Strategies (`--scaling`)

Outlier Clipping (`--clip-outliers`)

Split Strategies (`--time-column` & `--group-column`)