ValueError: could not convert string to float

The error#

ValueError: could not convert string to float: 'Mumbai'

The value after the colon changes: 'yes', '12,500', '₹499', '?', 'NA'. The cause is the same.

What it means#

Something tried to turn text into a number and failed. Models in scikit-learn, and NumPy's float conversions, work on numbers only. The message helpfully shows the exact value that refused to convert — read it. It tells you which of the two situations below you are in.

Why it happens#

Situation one: a genuine category column. 'Mumbai' or 'yes' is not a number and never will be. You passed a raw DataFrame with text columns straight into model.fit(X, y).

Situation two: numbers stored as text. '12,500' and '₹499' are numbers wearing costume — commas, currency symbols, percent signs, or placeholder strings like '?' for missing data. One such value makes pandas store the whole column as text.

How to fix it#

1. Find which columns are text.

print(X.select_dtypes(include="object").columns.tolist())

2. For category columns, encode them instead of dropping them. The clean way is a ColumnTransformer so the same encoding applies at training and prediction time:

from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OneHotEncoder
from sklearn.pipeline import make_pipeline
from sklearn.linear_model import LogisticRegression

cat_cols = ["city", "product"]
pre = ColumnTransformer(
    [("cat", OneHotEncoder(handle_unknown="ignore"), cat_cols)],
    remainder="passthrough",
)
model = make_pipeline(pre, LogisticRegression(max_iter=1000))
model.fit(X, y)

One-hot encoding turns each category into its own 0/1 column, which is what linear models and most others need.

3. For numbers-as-text, strip the costume and convert.

X["price"] = (
    X["price"].astype(str)
    .str.replace(",", "", regex=False)
    .str.replace("₹", "", regex=False)
)
X["price"] = pd.to_numeric(X["price"], errors="coerce")

errors="coerce" turns anything unconvertible into NaN. Count the NaNs afterwards — a large count means the column held something you did not expect.

4. Tell read_csv about placeholder strings up front.

df = pd.read_csv("data.csv", na_values=["?", "NA", "n/a", "-"])

Now placeholders arrive as proper missing values, and numeric columns stay numeric.

How to prevent it#

Run df.dtypes right after loading any file. A column you expected to be numeric showing object is this error waiting to happen. Handle categories with encoders inside a pipeline, not by hand, so training and inference always agree.

Related errors#

• ValueError: Input contains NaN, infinity or a value too large — the usual next error, after coercing to NaN
• ValueError: Unknown label type: 'continuous'
• ValueError: cannot convert float NaN to integer