feature pandas-dev#49580: support new-style float_format string in to_csv

Author: pedromfdiogo

pandas/io/formats/format.py

@@ -455,7 +455,7 @@ def __init__(
         self.na_rep = na_rep
         self.formatters = self._initialize_formatters(formatters)
         self.justify = self._initialize_justify(justify)
-        self.float_format = float_format
+        self.float_format = self._validate_float_format(float_format)
         self.sparsify = self._initialize_sparsify(sparsify)
         self.show_index_names = index_names
         self.decimal = decimal
@@ -850,6 +850,35 @@ def _get_column_name_list(self) -> list[Hashable]:
             names.append("" if columns.name is None else columns.name)
         return names
 
+    def _validate_float_format(
+        self, fmt: FloatFormatType | None
+    ) -> FloatFormatType | None:
+        """
+        Validates and processes the float_format argument.
+        Converts new-style format strings to callables.
+        """
+        
+        if fmt is None:
+            return None
+        
+        if callable(fmt):
+            return fmt
+        
+        if isinstance(fmt, str):
+            if "%" in fmt:
+                # Keeps old-style format strings as they are (C code handles them)
+                return fmt
+            else:
+        
+                try:
+                    _ = fmt.format(1.0)  # Test with an arbitrary float
+                    return lambda x: fmt.format(x)
+                except (ValueError, KeyError, IndexError) as e:
+                    
+                    raise ValueError(f"Invalid new-style format string {repr(fmt)}") from e
+                
+        # If fmt is neither None, nor callable, nor a successfully processed string,
+        raise ValueError("float_format must be a string or callable")
 
 class DataFrameRenderer:
     """Class for creating dataframe output in multiple formats.

pandas/tests/io/formats/test_csv_benchmarks.py

@@ -0,0 +1,21 @@
+import pytest
+import numpy as np
+import pandas as pd
+
+pytestmark = pytest.mark.usefixtures("benchmark")
+
+def test_benchmark_old_style_format(benchmark):
+    df = pd.DataFrame(np.random.rand(1000, 1000))
+    benchmark(lambda: df.to_csv(float_format="%.6f"))
+
+def test_benchmark_new_style_format(benchmark):
+    df = pd.DataFrame(np.random.rand(1000, 1000))
+    benchmark(lambda: df.to_csv(float_format="{:.6f}"))
+
+def test_benchmark_new_style_thousands(benchmark):
+    df = pd.DataFrame(np.random.rand(1000, 1000))
+    benchmark(lambda: df.to_csv(float_format="{:,.2f}"))
+
+def test_benchmark_callable_format(benchmark):
+    df = pd.DataFrame(np.random.rand(1000, 1000))
+    benchmark(lambda: df.to_csv(float_format=lambda x: f"{x:.6f}"))

pandas/tests/io/formats/test_to_csv.py

@@ -14,7 +14,7 @@
     compat,
 )
 import pandas._testing as tm
-
+import warnings
 
 class TestToCSV:
     def test_to_csv_with_single_column(self):
@@ -741,3 +741,124 @@ def test_to_csv_iterative_compression_buffer(compression):
             pd.read_csv(buffer, compression=compression, index_col=0), df
         )
         assert not buffer.closed
+
+
+def test_new_style_float_format_basic():
+    df = pd.DataFrame({"A": [1234.56789, 9876.54321]})
+    result = df.to_csv(float_format="{:.2f}")
+    expected = ",A\n0,1234.57\n1,9876.54\n"
+    assert result == expected
+
+def test_new_style_float_format_thousands():
+    df = pd.DataFrame({"A": [1234.56789, 9876.54321]})
+    result = df.to_csv(float_format="{:,.2f}")
+    expected = ',A\n0,"1,234.57"\n1,"9,876.54"\n'
+    assert result == expected
+
+def test_new_style_scientific_format():
+    df = pd.DataFrame({"A": [0.000123, 0.000456]})
+    result = df.to_csv(float_format="{:.2e}")
+    expected = ",A\n0,1.23e-04\n1,4.56e-04\n"
+    assert result == expected
+
+def test_new_style_with_nan():
+    df = pd.DataFrame({"A": [1.23, np.nan, 4.56]})
+    result = df.to_csv(float_format="{:.2f}", na_rep="NA")
+    expected = ",A\n0,1.23\n1,NA\n2,4.56\n"
+    assert result == expected
+
+def test_new_style_with_mixed_types():
+    df = pd.DataFrame({"A": [1.23, 4.56], "B": ["x", "y"]})
+    result = df.to_csv(float_format="{:.2f}")
+    expected = ",A,B\n0,1.23,x\n1,4.56,y\n"
+    assert result == expected
+
+def test_new_style_with_mixed_types_in_column():
+    df = pd.DataFrame({"A": [1.23, "text", 4.56]})
+    with warnings.catch_warnings(record=True) as w:
+        warnings.simplefilter("always")
+        result = df.to_csv(float_format="{:.2f}")
+        
+    expected = ",A\n0,1.23\n1,text\n2,4.56\n"
+    assert result == expected
+
+def test_invalid_new_style_format_missing_brace():
+    df = pd.DataFrame({"A": [1.23]})
+    with pytest.raises(ValueError, match="Invalid new-style format string '{:.2f"):
+        df.to_csv(float_format="{:.2f")
+
+def test_invalid_new_style_format_specifier():
+    df = pd.DataFrame({"A": [1.23]})
+    with pytest.raises(ValueError, match="Invalid new-style format string '{:.2z}'"):
+        df.to_csv(float_format="{:.2z}")
+
+def test_old_style_format_compatibility():
+    df = pd.DataFrame({"A": [1234.56789, 9876.54321]})
+    result = df.to_csv(float_format="%.2f")
+    expected = ",A\n0,1234.57\n1,9876.54\n"
+    assert result == expected
+
+def test_callable_float_format_compatibility():
+    df = pd.DataFrame({"A": [1234.56789, 9876.54321]})
+    result = df.to_csv(float_format=lambda x: f"{x:,.2f}")
+    expected = ',A\n0,"1,234.57"\n1,"9,876.54"\n'
+    assert result == expected
+
+def test_no_float_format():
+    df = pd.DataFrame({"A": [1.23, 4.56]})
+    result = df.to_csv(float_format=None)
+    expected = ",A\n0,1.23\n1,4.56\n"
+    assert result == expected
+
+def test_large_numbers():
+    df = pd.DataFrame({"A": [1e308, 2e308]})
+    result = df.to_csv(float_format="{:.2e}")
+    expected = ",A\n0,1.00e+308\n1,inf\n"
+    assert result == expected
+
+def test_zero_and_negative():
+    df = pd.DataFrame({"A": [0.0, -1.23456]})
+    result = df.to_csv(float_format="{:+.2f}")
+    expected = ",A\n0,+0.00\n1,-1.23\n"
+    assert result == expected
+
+def test_unicode_format():
+    df = pd.DataFrame({"A": [1.23, 4.56]})
+    result = df.to_csv(float_format="{:.2f}€", encoding="utf-8")
+    expected = ",A\n0,1.23€\n1,4.56€\n"
+    assert result == expected
+
+def test_empty_dataframe():
+    df = pd.DataFrame({"A": []})
+    result = df.to_csv(float_format="{:.2f}")
+    expected = ",A\n"
+    assert result == expected
+
+def test_multi_column_float():
+    df = pd.DataFrame({"A": [1.23, 4.56], "B": [7.89, 0.12]})
+    result = df.to_csv(float_format="{:.2f}")
+    expected = ",A,B\n0,1.23,7.89\n1,4.56,0.12\n"
+    assert result == expected
+
+def test_invalid_float_format_type():
+    df = pd.DataFrame({"A": [1.23]})
+    with pytest.raises(ValueError, match="float_format must be a string or callable"):
+        df.to_csv(float_format=123)
+
+def test_new_style_with_inf():
+    df = pd.DataFrame({"A": [1.23, np.inf, -np.inf]})
+    result = df.to_csv(float_format="{:.2f}", na_rep="NA")
+    expected = ",A\n0,1.23\n1,inf\n2,-inf\n"
+    assert result == expected
+
+def test_new_style_with_precision_edge():
+    df = pd.DataFrame({"A": [1.23456789]})
+    result = df.to_csv(float_format="{:.10f}")
+    expected = ",A\n0,1.2345678900\n"
+    assert result == expected
+
+def test_new_style_with_template():
+    df = pd.DataFrame({"A": [1234.56789]})
+    result = df.to_csv(float_format="Value: {:,.2f}")
+    expected = ',A\n0,"Value: 1,234.57"\n'
+    assert result == expected