29:null 2a:["$","title",null,{"children":"Synthetic data generation in Python | dduyg | Observable"}] 2f:Tbc0,hljs = { await new Promise((resolve, reject) => { const s = document.createElement("script"); s.src = "https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/highlight.min.js"; s.onload = resolve; s.onerror = reject; document.head.appendChild(s); }); await new Promise((resolve, reject) => { const s = document.createElement("script"); s.src = "https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/languages/python.min.js"; s.onload = resolve; s.onerror = reject; document.head.appendChild(s); }); const style = document.createElement("style"); style.textContent = ` .ll-pre { margin: 0 !important; padding: 0 !important; background: transparent !important; border: 1px solid #e2e2e2 !important; border-radius: 6px !important; overflow-x: auto !important; white-space: pre !important; max-width: 100% !important; box-sizing: border-box !important; } .ll-pre code.hljs { display: block !important; background: #FBFBFB !important; color: #333333 !important; padding: 16px 20px !important; font-size: 12.5px !important; line-height: 1.6 !important; font-family: 'JetBrains Mono', 'Fira Code', 'Cascadia Code', monospace !important; white-space: pre !important; overflow-x: visible !important; border-radius: 6px !important; } .ll-pre .hljs-comment, .ll-pre .hljs-quote { color: #797979 !important; font-style: italic; } .ll-pre .hljs-keyword { color: #7A5A3B !important; font-weight: 700; } .ll-pre .hljs-title, .ll-pre .hljs-function { color: #2A8099 !important; font-weight: 700; } .ll-pre .hljs-string, .ll-pre .hljs-attr { color: #478309 !important; font-weight: 700; } .ll-pre .hljs-number, .ll-pre .hljs-literal { color: #D62888 !important; font-weight: 700; } .ll-pre .hljs-operator, .ll-pre .hljs-punctuation { color: #333333 !important; } .ll-pre .hljs-built_in, .ll-pre .hljs-builtin-name { color: #0B7BA9 !important; font-weight: 700; } .ll-pre .hljs-variable, .ll-pre .hljs-name { color: #333333 !important; } .ll-pre .hljs-deletion, .ll-pre .hljs-error { color: #fff !important; background: #D62888 !important; font-weight: bold; } .ll-block { max-width: 860px; margin: 0 0 2.5rem; font-family: "Raleway", sans-serif; box-sizing: border-box; } .ll-label { font-size: 10px; letter-spacing: 0.15em; text-transform: uppercase; color: #797979; margin-bottom: 4px; font-family: monospace; } .ll-title { font-size: 1.1rem; font-weight: bold; font-family: "Averia Serif Libre", serif; color: #2A8099; margin: 0 0 6px; } .ll-desc { font-size: 0.85rem; line-height: 1.75; margin: 0 0 10px; } `; document.head.appendChild(style); return window.hljs; }30:[] 2d:{"id":"0ed372515d882db8","slug":"synthetic-data-pipeline-in-python","title":"Synthetic data generation in Python","thumbnail":"e5a5e46d31bf1353f411216c0ed527c0ead545ccc19da70df52c5b49e84f2b7f","access_level":"public","create_time":"$D2026-06-02T19:46:44.016Z","update_time":"$D2026-06-04T20:00:44.775Z","view_time":null,"first_public_time":"$D2026-06-02T19:46:44.016Z","has_unpublished_edits":null,"fork":false,"stars":0,"starred":false,"owner":{"type":"team","id":"1fa0c77b14fd90d0","login":"dduyg","name":"","avatar_url":"https://avatars-next.observableusercontent.com/8922701a2d973be0158565df96ce0b2fe9ffd55df053350d59973a811bac06af","tier":"starter_2024"},"version":77,"publish_version":null,"role":null,"authored":false,"authors":[{"type":"individual","id":"248f116c85e550e3","login":"dduyg","name":"Duygu Dağdelen","avatar_url":"https://avatars-next.observableusercontent.com/ca098eb30fb37c129368c39d70e98a9a8444cfaa7496c72781905ac410af8cad","tier":null}],"members":[{"type":"team","id":"1fa0c77b14fd90d0","login":"dduyg","name":"","avatar_url":"https://avatars-next.observableusercontent.com/8922701a2d973be0158565df96ce0b2fe9ffd55df053350d59973a811bac06af","tier":"starter_2024","role":"editor"},{"type":"individual","id":"248f116c85e550e3","login":"dduyg","name":"Duygu Dağdelen","avatar_url":"https://avatars-next.observableusercontent.com/ca098eb30fb37c129368c39d70e98a9a8444cfaa7496c72781905ac410af8cad","tier":null,"role":"editor"}],"comments":[],"custom_thumbnail":true,"trashed":false,"lapsed":null,"subscribed":null,"forks":0,"collection":null,"type":"notebook","body":{"title":"Synthetic data generation in Python","cells":[{"id":0,"value":"# Synthetic data generation in Python\n#### using *SDV GaussianCopula* + *ydata-profiling*.\n\n
\n\n---\n","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":5,"value":"codeBlock(\n \"01 / Imports\",\n \"Library Imports\",\n \"Import pandas for DataFrames, GaussianCopula from SDV to model joint distributions of real data and sample synthetic rows, and ProfileReport from ydata-profiling to audit data health before synthesis begins.\",\n `import pandas as pd\nfrom sdv.tabular import GaussianCopula\nfrom ydata_profiling import ProfileReport`\n)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":37,"value":"
","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":8,"value":"codeBlock(\n \"02 / Load & Profile\",\n \"Load Dataset + Generate Profiling Report\",\n \"Read the raw CSV into a DataFrame, then fire off a full HTML profiling report. This surfaces distributions, missing value patterns, correlations, and outliers — critical intelligence before training any synthesis model.\",\n `# Load real dataset\ndf = pd.read_csv(\"raw_data.csv\")\n\n# Generate profiling report\nprofile = ProfileReport(df, title=\"Data Health Check\")\nprofile.to_file(\"report.html\")`\n)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":42,"value":"
","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":11,"value":"codeBlock(\n \"03 / Core Function\",\n \"Train GaussianCopula & Generate Synthetic Samples\",\n \"GaussianCopula learns the full joint distribution of your columns — correlations included — and samples from it. After generation, float columns are rounded to 2 decimal places to prevent floating-point noise from inflating file size.\",\n `def generate_synthetic_data(real_data: pd.DataFrame, n_samples: int = 1000):\n \"\"\"\n Train GaussianCopula model from SDV and generate synthetic samples.\n \"\"\"\n print(\"Training SDV model...\")\n model = GaussianCopula()\n model.fit(real_data) # fit model to real data\n\n print(f\"Generating {n_samples} synthetic samples...\")\n synthetic_data = model.sample(n_samples)\n\n # Post-processing: round numeric columns if needed\n for col in synthetic_data.select_dtypes(include='float').columns:\n synthetic_data[col] = synthetic_data[col].round(2)\n\n return synthetic_data`\n)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":46,"value":"
","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":14,"value":"codeBlock(\n \"04 / Run & Export\",\n \"Generate 500 Synthetic Rows & Save to CSV\",\n \"We call the function requesting 500 rows. SDV doesn't need retraining to change the sample count — the fitted model can be sampled repeatedly. The result is saved to CSV for downstream use.\",\n `# Generate 500 synthetic rows\nsynthetic_df = generate_synthetic_data(df, n_samples=500)\n\n# Save synthetic data to CSV\noutput_file = \"synthetic_data.csv\"\nsynthetic_df.to_csv(output_file, index=False)\nprint(f\"Synthetic data saved to {output_file}\")`\n)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":50,"value":"
","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":17,"value":"codeBlock(\n \"05 / Inspect & Validate\",\n \"Preview Rows, Loop Columns, Check for Nulls\",\n \"Three quick validation steps: row-by-row dict preview for spot-checking, a column-count loop, and a null guard. Null checking matters — GaussianCopula can introduce NaNs when modelling sparse or highly skewed columns.\",\n `# Optional: loop to inspect first 5 rows\nprint(\"\\\\nPreview of synthetic data:\")\nfor i, row in synthetic_df.head(5).iterrows():\n print(row.to_dict())\n\n# Conditional: check for nulls\nif synthetic_df.isnull().sum().sum() > 0:\n print(\"Warning: Synthetic data contains nulls!\")\nelse:\n print(\"No missing values in synthetic data.\")\n\n# Log column counts\nnum_samples = 10\nfor i in range(num_samples):\n print(f\"Sample {i+1} has {len(synthetic_df.columns)} columns\")`\n)","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":20,"value":"

\n\n---\n### /imports","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":23,"value":"$2f","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":26,"value":"function codeBlock(label, title, desc, code, lang = \"python\") {\n // Referencing `hljs` here tells Observable's runtime to wait for Cell 1\n const highlighted = hljs.highlight(code.trim(), { language: lang }).value;\n\n const wrap = document.createElement(\"div\");\n wrap.className = \"ll-block\";\n wrap.innerHTML = `\n

${label}

${title}

${desc}

${highlighted}

\n `;\n return wrap;\n}","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"}],"files":"$W30","stdlib":"1","titling":"auto"}} 2e:"$2d:body"