DataStore Pandas compatibility

DataStore implements 209 pandas DataFrame methods for full API compatibility. Your existing pandas code works with minimal changes.

Compatibility Approach

# Typical migration - just change the import
- import pandas as pd
+ from chdb import datastore as pd

# Your code works unchanged
df = pd.read_csv("data.csv")
result = df[df['age'] > 25].groupby('city')['salary'].mean()

Key principles:

All 209 pandas DataFrame methods implemented
Lazy evaluation for SQL optimization
Automatic type wrapping (DataFrame → DataStore, Series → ColumnExpr)
Immutable operations (no inplace=True)

Attributes and Properties

Property	Description	Triggers Execution
`shape`	(rows, columns) tuple	Yes
`columns`	Column names (Index)	Yes
`dtypes`	Column data types	Yes
`values`	NumPy array	Yes
`index`	Row index	Yes
`size`	Number of elements	Yes
`ndim`	Number of dimensions	No
`empty`	Is DataFrame empty	Yes
`T`	Transpose	Yes
`axes`	List of axes	Yes

Examples:

from chdb import datastore as pd

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

print(ds.shape)      # (1000, 5)
print(ds.columns)    # Index(['name', 'age', 'city', 'salary', 'dept'])
print(ds.dtypes)     # name: object, age: int64, ...
print(ds.empty)      # False

Indexing and Selection

Method	Description	Example
`df['col']`	Select column	`ds['age']`
`df[['col1', 'col2']]`	Select columns	`ds[['name', 'age']]`
`df[condition]`	Boolean indexing	`ds[ds['age'] > 25]`
`df.loc[...]`	Label-based access	`ds.loc[0:10, 'name']`
`df.iloc[...]`	Integer-based access	`ds.iloc[0:10, 0:3]`
`df.at[...]`	Single value by label	`ds.at[0, 'name']`
`df.iat[...]`	Single value by position	`ds.iat[0, 0]`
`df.head(n)`	First n rows	`ds.head(10)`
`df.tail(n)`	Last n rows	`ds.tail(10)`
`df.sample(n)`	Random sample	`ds.sample(100)`
`df.select_dtypes()`	Select by Dtype	`ds.select_dtypes(include='number')`
`df.query()`	Query expression	`ds.query('age > 25')`
`df.where()`	Conditional replace	`ds.where(ds['age'] > 0, 0)`
`df.mask()`	Inverse where	`ds.mask(ds['age'] < 0, 0)`
`df.isin()`	Value membership	`ds['city'].isin(['NYC', 'LA'])`
`df.get()`	Safe column access	`ds.get('col', default=None)`
`df.xs()`	Cross-section	`ds.xs('key')`
`df.pop()`	Remove column	`ds.pop('col')`

Statistical Methods

Method	Description	SQL Equivalent
`mean()`	Mean value	`AVG()`
`median()`	Median value	`MEDIAN()`
`mode()`	Mode value	-
`std()`	Standard deviation	`STDDEV()`
`var()`	Variance	`VAR()`
`min()`	Minimum	`MIN()`
`max()`	Maximum	`MAX()`
`sum()`	Sum	`SUM()`
`prod()`	Product	-
`count()`	Non-null count	`COUNT()`
`nunique()`	Unique count	`UNIQ()`
`value_counts()`	Value frequencies	`GROUP BY`
`quantile()`	Quantile	`QUANTILE()`
`describe()`	Summary statistics	-
`corr()`	Correlation matrix	`CORR()`
`cov()`	Covariance matrix	`COV()`
`corrwith()`	Pairwise correlation	-
`rank()`	Rank values	`RANK()`
`abs()`	Absolute values	`ABS()`
`round()`	Round values	`ROUND()`
`clip()`	Clip values	-
`cumsum()`	Cumulative sum	Window function
`cumprod()`	Cumulative product	Window function
`cummin()`	Cumulative min	Window function
`cummax()`	Cumulative max	Window function
`diff()`	Difference	Window function
`pct_change()`	Percent change	Window function
`skew()`	Skewness	`SKEW()`
`kurt()`	Kurtosis	`KURT()`
`sem()`	Standard error	-
`all()`	All true	-
`any()`	Any true	-
`idxmin()`	Index of min	-
`idxmax()`	Index of max	-

Examples:

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

# Basic statistics
print(ds['salary'].mean())
print(ds['age'].std())
print(ds.describe())

# Group statistics
print(ds.groupby('department')['salary'].mean())
print(ds.groupby('city').agg({'salary': ['mean', 'std'], 'age': 'count'}))

Data Manipulation

Method	Description
`drop()`	Drop rows/columns
`drop_duplicates()`	Remove duplicates
`duplicated()`	Mark duplicates
`dropna()`	Remove missing values
`fillna()`	Fill missing values
`ffill()`	Forward fill
`bfill()`	Backward fill
`interpolate()`	Interpolate values
`replace()`	Replace values
`rename()`	Rename columns/index
`rename_axis()`	Rename axis
`assign()`	Add new columns
`astype()`	Convert types
`convert_dtypes()`	Infer types
`copy()`	Copy DataFrame

Examples:

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

# Drop operations
result = ds.drop(columns=['unused_col'])
result = ds.drop_duplicates(subset=['user_id'])
result = ds.dropna(subset=['email'])

# Fill operations
result = ds.fillna(0)
result = ds.fillna({'age': 0, 'name': 'Unknown'})

# Transform operations
result = ds.rename(columns={'old_name': 'new_name'})
result = ds.assign(
    full_name=lambda x: x['first_name'] + ' ' + x['last_name'],
    age_group=lambda x: pd.cut(x['age'], bins=[0, 25, 50, 100])
)

Sorting and Ranking

Method	Description
`sort_values()`	Sort by values
`sort_index()`	Sort by index
`nlargest()`	N largest values
`nsmallest()`	N smallest values

Examples:

# Sort by single column
result = ds.sort_values('salary', ascending=False)

# Sort by multiple columns
result = ds.sort_values(['department', 'salary'], ascending=[True, False])

# Get top/bottom N
result = ds.nlargest(10, 'salary')
result = ds.nsmallest(5, 'age')

Reshaping

Method	Description
`pivot()`	Pivot table
`pivot_table()`	Pivot with aggregation
`melt()`	Unpivot
`stack()`	Stack columns to index
`unstack()`	Unstack index to columns
`transpose()` / `T`	Transpose
`explode()`	Explode lists to rows
`squeeze()`	Reduce dimensions
`droplevel()`	Drop index level
`swaplevel()`	Swap index levels
`reorder_levels()`	Reorder levels

Examples:

# Pivot table
result = ds.pivot_table(
    values='amount',
    index='region',
    columns='product',
    aggfunc='sum'
)

# Melt (unpivot)
result = ds.melt(
    id_vars=['name'],
    value_vars=['score1', 'score2', 'score3'],
    var_name='test',
    value_name='score'
)

# Explode arrays
result = ds.explode('tags')

Combining / Joining

Method	Description
`merge()`	SQL-style merge
`join()`	Join on index
`concat()`	Concatenate
`append()`	Append rows
`combine()`	Combine with function
`combine_first()`	Combine with priority
`update()`	Update values
`compare()`	Show differences

Examples:

# Merge (join)
result = pd.merge(df1, df2, on='id', how='left')
result = df1.join(df2, on='id')

# Concatenate
result = pd.concat([df1, df2, df3])
result = pd.concat([df1, df2], axis=1)

Binary Operations

Method	Description
`add()` / `radd()`	Addition
`sub()` / `rsub()`	Subtraction
`mul()` / `rmul()`	Multiplication
`div()` / `rdiv()`	Division
`truediv()` / `rtruediv()`	True division
`floordiv()` / `rfloordiv()`	Floor division
`mod()` / `rmod()`	Modulo
`pow()` / `rpow()`	Power
`dot()`	Matrix multiplication

Examples:

# Arithmetic operations
result = ds['col1'].add(ds['col2'])
result = ds['price'].mul(ds['quantity'])

# With fill_value for missing data
result = ds['col1'].add(ds['col2'], fill_value=0)

Comparison Operations

Method	Description
`eq()`	Equal
`ne()`	Not equal
`lt()`	Less than
`le()`	Less than or equal
`gt()`	Greater than
`ge()`	Greater than or equal
`equals()`	Test equality
`compare()`	Show differences

Function Application

Method	Description
`apply()`	Apply function
`applymap()`	Apply element-wise
`map()`	Map values
`agg()` / `aggregate()`	Aggregate
`transform()`	Transform
`pipe()`	Pipe functions
`groupby()`	Group by

Examples:

# Apply function
result = ds['name'].apply(lambda x: x.upper())
result = ds.apply(lambda row: row['a'] + row['b'], axis=1)

# Aggregate
result = ds.agg({'col1': 'sum', 'col2': 'mean'})
result = ds.agg(['sum', 'mean', 'std'])

# Pipe
result = (ds
    .pipe(filter_active)
    .pipe(calculate_metrics)
    .pipe(format_output)
)

Time Series

Method	Description
`rolling()`	Rolling window
`expanding()`	Expanding window
`ewm()`	Exponentially weighted
`resample()`	Resample time series
`shift()`	Shift values
`asfreq()`	Convert frequency
`asof()`	Latest value as of
`at_time()`	Select at time
`between_time()`	Select time range
`first()` / `last()`	First/last periods
`to_period()`	Convert to period
`to_timestamp()`	Convert to timestamp
`tz_convert()`	Convert timezone
`tz_localize()`	Localize timezone

Examples:

# Rolling window
result = ds['value'].rolling(window=7).mean()

# Expanding window
result = ds['value'].expanding().sum()

# Shift
result = ds['value'].shift(1)  # Lag
result = ds['value'].shift(-1)  # Lead

Missing Data

Method	Description
`isna()` / `isnull()`	Detect missing
`notna()` / `notnull()`	Detect non-missing
`dropna()`	Drop missing
`fillna()`	Fill missing
`ffill()`	Forward fill
`bfill()`	Backward fill
`interpolate()`	Interpolate
`replace()`	Replace values

I/O Methods

Method	Description
`to_csv()`	Export to CSV
`to_json()`	Export to JSON
`to_excel()`	Export to Excel
`to_parquet()`	Export to Parquet
`to_feather()`	Export to Feather
`to_sql()`	Export to SQL database
`to_pickle()`	Pickle
`to_html()`	HTML table
`to_latex()`	LaTeX table
`to_markdown()`	Markdown table
`to_string()`	String representation
`to_dict()`	Dictionary
`to_records()`	Records
`to_numpy()`	NumPy array
`to_clipboard()`	Clipboard

See I/O Operations for detailed documentation.

Iteration

Method	Description
`items()`	Iterate (column, Series)
`iterrows()`	Iterate (index, Series)
`itertuples()`	Iterate as named tuples

Key Differences from Pandas

1. Return Types

# Pandas returns Series
pdf['col']  # → pd.Series

# DataStore returns ColumnExpr (lazy)
ds['col']   # → ColumnExpr

2. Lazy Execution

# DataStore operations are lazy
result = ds.filter(ds['age'] > 25)  # Not executed yet
df = result.to_df()  # Executed here

3. No inplace Parameter

# Pandas
df.drop(columns=['col'], inplace=True)

# DataStore (always returns new object)
ds = ds.drop(columns=['col'])

4. Comparing Results

# Use to_pandas() for comparison
pd.testing.assert_frame_equal(
    ds.to_pandas(),
    expected_df
)

See Key Differences for complete details.