Pandas新增列类型问题：event_id为何转为float及解决方法

Great question! Let's break down why your event_id values (180, 181) are showing up as floats (180.0, 181.0) and how to fix it.

Root Cause #

The issue comes down to how pandas handles dynamic column addition and type inference with empty DataFrames:

1. You start with data_for_csv = pd.DataFrame(columns=["ds", "yhat", "yhat_lower", "yhat_upper"]), an empty DataFrame that doesn't include the event_id column.
2. When you append rows containing the event_id column, pandas has to add this column on the fly.
3. Since the initial DataFrame was empty, pandas prioritizes flexibility over strict typing: it defaults to float for the new column because float can accommodate missing values (NaN) whereas integer types cannot. Even though your event_id values are integers, pandas assumes the column might eventually have missing data and converts it to float as a precaution.

In some cases, if other columns in your forecast DataFrame are float types, this can reinforce pandas' decision to cast event_id to float to maintain consistency, but this is secondary to the empty DataFrame issue.

Fixes to Keep event_id as Integer #

1. Define event_id upfront with explicit dtype #

Create your initial DataFrame with the event_id column included and specify its integer type from the start. This tells pandas exactly what type to use, eliminating implicit casting:

# Include event_id in columns and set its dtype explicitly
data_for_csv = pd.DataFrame(
    columns=["event_id", "ds", "yhat", "yhat_lower", "yhat_upper"],
    dtype={"event_id": int}
)

2. Convert the column to integer after appending #

If you want to keep your original initial DataFrame setup, you can cast the event_id column back to integer once all appends are complete. Just ensure there are no missing values (NaN) in the column first (which there aren't in your case):

# After all appends finish
data_for_csv["event_id"] = data_for_csv["event_id"].astype(int)

3. Enforce integer type when adding event_id to forecast #

When adding the event_id column to the forecast DataFrame, explicitly set its type to integer to avoid any pre-append casting:

# Explicitly create an integer series for event_id
forecast["event_id"] = pd.Series([event_id] * len(forecast), dtype=int)

Verification #

After implementing any fix, check the column types to confirm event_id is now integer:

print(data_for_csv.dtypes)

You should see output like this:

event_id        int64
ds             object
yhat          float64
yhat_lower    float64
yhat_upper    float64
dtype: object

内容的提问来源于stack exchange，提问作者Joey Coder