Different 'drop_invalid_rows=True' behavior on similar DataFrame schemas

[Nico-VC]

Hi!

I am trying to find a workaround for the behavior of drop_invalid_rows=True when passing a dtype to a mixed string-int column.

import pandera as pa

from pandera import Column, DataFrameSchema, Check, Index

df = pd.DataFrame({"counter": ["a", "2", "3"]})

schema = DataFrameSchema(
    {
        "counter": Column(
                    int,
                    checks=None,
                    coerce=True
                    )
    },
    drop_invalid_rows=True
)
df_val = schema.validate(df, lazy=True)

Setting drop_invalid_rows at a DataFrame Schema level, the validated 'df_val' returns type 'object' and drops the row with "a" value.

Whereas, setting drop_invalid_rows=True at a Column level returns schema error for "a" value as a failure case.

import pandera as pa

from pandera import Column, DataFrameSchema, Check, Index

df = pd.DataFrame({"counter": ["a", "2", "3"]})


schema = DataFrameSchema(
    {
        "counter": Column(
                    int,
                    checks=None,
                    coerce=True,
                    drop_invalid_rows=True
                    )
    },
)
df_val = schema.validate(df, lazy=True)

---
    raise SchemaErrors(
pandera.errors.SchemaErrors: Schema None: A total of 1 schema errors were found.
Error Counts
------------
- SchemaErrorReason.SCHEMA_COMPONENT_CHECK: 1
Schema Error Summary
--------------------
                                             failure_cases  n_failure_cases
schema_context column  check                                               
Column         counter coerce_dtype('int64')           [a]                1

How come there is different output behavior for these similar schemas?

The first case makes sense. However is there a workaround to coerce to integer type instead of getting object type, with the invalid row being dropped?