Repository navigation
Misleading error message when an invalid y column name is provided to px.line #4040
Description
When attempting to plot multiple y values using px.line (or similar), erroneous column names lead to a misleading error message that states that the length of the arguments is incorrect, when in fact the problem is that there the column name is wrong.
Simple to reproduce:
import plotly.express as px
import pandas as pd
import numpy as np
x = np.linspace(0, 10, 100)
data1 = x**2
data2 = x**3
df = pd.DataFrame({"x": x, "data1": data1, "data2": data2})
px.line(df, x = "x", y = ["data1", "data2"]) # works fine
px.line(df, x = "x", y = ["data1", "Data2"])
# this last line gives a misleading error message:
# ValueError: All arguments should have the same length. The length of argument `y` is 2,
# whereas the length of previously-processed arguments ['x'] is 100
Activity
nicolaskruchten commentedon Feb 11, 2023
Yes, we should definitely find a way to give you a better hint there that there's a typo in the contents of
y. It's a bit tricky to do and we're love some help to figure out how to do it! The relevant code is near here: https://github.com/plotly/plotly.py/blob/master/packages/python/plotly/plotly/express/_core.py#L1247nicolaskruchten commentedon Feb 11, 2023
The reason it's tricky is that we land deep in that
ifblock based on some determination that happened way ahead of time here https://github.com/plotly/plotly.py/blob/master/packages/python/plotly/plotly/express/_core.py#L1315 ... basically up front PX decides if it's in "wide mode" based on whether or not the contents ofyare all column names, and then much later if they're not, it checks the length of the dataypoints to against the length of the dataxpoints to.nicolaskruchten commentedon Feb 11, 2023
See also #3359 and #3474 and #2586
abarpan3 commentedon Jul 14, 2023
Hi, was exploring this issue a bit. So the build_dataframe function has another function is_col_list to check if the supplied argument for x/y is a list or not (hence wide-form) https://github.com/plotly/plotly.py/blob/master/packages/python/plotly/plotly/express/_core.py#L1339
Now this is_col_list function has a check inside it to see if a column is present in the dataframe, and if not it returns wide_mode as false.
https://github.com/plotly/plotly.py/blob/master/packages/python/plotly/plotly/express/_core.py#L1036
Here, in the is_col_list functionwe can catch the erroneous column name and give the user a more helpful error message. So I tried to make 2 changes,
Line 1035 : Checking if y value is of pd.Series type and returning wide_mode as false, if the condition is fulfilled. This takes care of any direct array inputs in x/y arguments like below example.
px.scatter(df1,x = df1.sales, y = df2.market)**Line 1042 - 1048 ** : Checking additionally if col_name supplied in list is present in df. And returning a more helpful error message. Attached image shows both changes
Could anyone kindly provide any feedback if this seems like the right approach?
Improve px error for unknown column names in a list passed to x or y