def marshall_data_frame(data, proto_df):
"""Convert a pandas.DataFrame into a proto.DataFrame.
Parameters
----------
data : pandas.DataFrame, numpy.ndarray, Iterable, dict, DataFrame, Styler, or None
Something that is or can be converted to a dataframe.
proto_df : proto.DataFrame
Output. The protobuf for a Streamlit DataFrame proto.
"""
if isinstance(data, pa.Table):
raise errors.StreamlitAPIException("""
pyarrow tables are not supported by Streamlit's legacy DataFrame serialization (i.e. with `config.dataFrameSerialization = "legacy"`).
To be able to use pyarrow tables, please enable pyarrow by changing the config setting,
`config.dataFrameSerialization = "arrow"`
""")
df = type_util.convert_anything_to_df(data)
# Convert df into an iterable of columns (each of type Series).
df_data = (df.iloc[:, col] for col in range(len(df.columns)))
_marshall_table(df_data, proto_df.data)
_marshall_index(df.columns, proto_df.columns)
_marshall_index(df.index, proto_df.index)
styler = data if type_util.is_pandas_styler(data) else None
_marshall_styles(proto_df.style, df, styler)
def generate_chart(chart_type, data, width=0, height=0):
if data is None:
# Use an empty-ish dict because if we use None the x axis labels rotate
# 90 degrees. No idea why. Need to debug.
data = {"": []}
if isinstance(data, pa.Table):
raise errors.StreamlitAPIException(
"""
pyarrow tables are not supported by Streamlit's legacy DataFrame serialization (i.e. with `config.dataFrameSerialization = "legacy"`).
To be able to use pyarrow tables, please enable pyarrow by changing the config setting,
`config.dataFrameSerialization = "arrow"`
"""
)
if not isinstance(data, pd.DataFrame):
data = type_util.convert_anything_to_df(data)
index_name = data.index.name
if index_name is None:
index_name = "index"
data = pd.melt(data.reset_index(), id_vars=[index_name])
if chart_type == "area":
opacity = {"value": 0.7}
else:
opacity = {"value": 1.0}
# Set the X and Y axes' scale to "utc" if they contain date values.
# This causes time data to be displayed in UTC, rather the user's local
# time zone. (By default, vega-lite displays time data in the browser's
# local time zone, regardless of which time zone the data specifies:
# https://vega.github.io/vega-lite/docs/timeunit.html#output).
x_scale = (
alt.Scale(type="utc") if _is_date_column(data, index_name) else alt.Undefined
)
y_scale = alt.Scale(type="utc") if _is_date_column(data, "value") else alt.Undefined
x_type = alt.Undefined
# Bar charts should have a discrete (ordinal) x-axis, UNLESS type is date/time
# https://github.com/streamlit/streamlit/pull/2097#issuecomment-714802475
if chart_type == "bar" and not _is_date_column(data, index_name):
x_type = "ordinal"
chart = (
getattr(alt.Chart(data, width=width, height=height), "mark_" + chart_type)()
.encode(
alt.X(index_name, title="", scale=x_scale, type=x_type),
alt.Y("value", title="", scale=y_scale),
alt.Color("variable", title="", type="nominal"),
alt.Tooltip([index_name, "value", "variable"]),
opacity=opacity,
)
.interactive()
)
return chart
def convert_anything_to_df(df):
"""Try to convert different formats to a Pandas Dataframe.
Parameters
----------
df : ndarray, Iterable, dict, DataFrame, Styler, pa.Table, None, dict, list, or any
Returns
-------
pandas.DataFrame
"""
# This is inefficent as the data will be converted back to Arrow
# when marshalled to protobuf, but area/bar/line charts need
# DataFrame magic to generate the correct output.
if isinstance(df, pa.Table):
return df.to_pandas()
if is_type(df, _PANDAS_DF_TYPE_STR):
return df
if is_pandas_styler(df):
return df.data
import pandas as pd
if is_type(df, "numpy.ndarray") and len(df.shape) == 0:
return pd.DataFrame([])
# Try to convert to pandas.DataFrame. This will raise an error is df is not
# compatible with the pandas.DataFrame constructor.
try:
return pd.DataFrame(df)
except ValueError:
raise errors.StreamlitAPIException("""
Unable to convert object of type `%(type)s` to `pandas.DataFrame`.
Offending object:
```py
%(object)s
```""" % {
"type": type(df),
"object": df,
})
def convert_anything_to_df(df):
"""Try to convert different formats to a Pandas Dataframe.
Parameters
----------
df : ndarray, Iterable, dict, DataFrame, Styler, None, dict, list, or any
Returns
-------
pandas.DataFrame
"""
if is_type(df, _PANDAS_DF_TYPE_STR):
return df
if is_pandas_styler(df):
return df.data
import pandas as pd
if is_type(df, "numpy.ndarray") and len(df.shape) == 0:
return pd.DataFrame([])
# Try to convert to pandas.DataFrame. This will raise an error is df is not
# compatible with the pandas.DataFrame constructor.
try:
return pd.DataFrame(df)
except ValueError:
raise errors.StreamlitAPIException("""
Unable to convert object of type `%(type)s` to `pandas.DataFrame`.
Offending object:
```py
%(object)s
```""" % {
"type": type(df),
"object": df,
})
def data_frame_to_bytes(df: DataFrame) -> bytes:
"""Serialize pandas.DataFrame to bytes using Apache Arrow.
Parameters
----------
df : pandas.DataFrame
A dataframe to convert.
"""
try:
table = pa.Table.from_pandas(df)
return pyarrow_table_to_bytes(table)
except Exception as e:
_NUMPY_DTYPE_ERROR_MESSAGE = "Could not convert dtype"
if _NUMPY_DTYPE_ERROR_MESSAGE in str(e):
raise errors.NumpyDtypeException("""
Unable to convert `numpy.dtype` to `pyarrow.DataType`.
This is likely due to a bug in Arrow (see https://issues.apache.org/jira/browse/ARROW-14087).
As a temporary workaround, you can convert the DataFrame cells to strings with `df.astype(str)`.
""")
else:
raise errors.StreamlitAPIException(e)