I'm getting the following warning after upgrading to version 1.0 of scikit-learn:
UserWarning: X does not have valid feature names, but IsolationForest was fitted with feature name
I cannot find in the docs on what is a "valid feature name". How do I deal with this warning?
57 Answers
I got the same warning message with another sklearn model. I realized that it was showing up because I fitted the model with a data in a dataframe, and then used only the values to predict. From the moment I fixed that, the warning disappeared.
Here is an example:
model_reg.fit(scaled_x_train, y_train[vp].values)
data_pred = model_reg.predict(scaled_x_test.values)This first code had the warning, because scaled_x_train is a DataFrame with feature names, while scaled_x_test.values is only values, without feature names. Then, I changed to this:
model_reg.fit(scaled_x_train.values, y_train[vp].values)
data_pred = model_reg.predict(scaled_x_test.values)And now there are no more warnings on my code.
4I was getting very similar error but on module DecisionTreeClassifier for Fit and Predict.
Initially I was sending dataframe as input to fit with headers and I got the error.
When I trimmed to remove the headers and sent only values then the error got disappeared. Sample code before and after changes.
Code with Warning:
model = DecisionTreeClassifier()
model.fit(x,y) #Here x includes the dataframe with headers
predictions = model.predict([ [20,1], [20,0]
])
print(predictions)Code without Warning:
model = DecisionTreeClassifier()
model.fit(x.values,y) #Here x.values will have only values without headers
predictions = model.predict([ [20,1], [20,0]
])
print(predictions)I had also the same problem .The problem was due to fact that I fitted the model with X train data as dataframe (model.fit(X,Y)) and I make a prediction with with X test as an array ( model.predict([ [20,0] ]) ) . To solve that I have converted the X train dataframe into an array as illustrated bellow .
BEFORE
model = DecisionTreeClassifier()
model.fit(X,Y) # X train here is a dataFrame
predictions = model.predict([20,0]) ## generates warning AFTER
model = DecisionTreeClassifier()
X = X.values # conversion of X into array
model.fit(X,Y)
model.predict([ [20,0] ]) #now ok , no warningThe other answers so far recommend (re)training using a numpy array instead of a dataframe for the training data. The warning is a sort of safety feature, to ensure you're passing the data you meant to, so I would suggest to pass a dataframe (with correct column labels!) to the predict function instead, e.g.:
test_row = pd.Dataframe({ "<feat1_name>": [20], "<feat2_name>": [0],
})
model.predict(test_row)Also, note that it's just a warning, not an error. You can ignore the warning and proceed with the rest of your code without problem; just be sure that the data is in the same order as it was trained with!
1Everyone seems to be suggesting .values; however, say, you don't want to modify your training process or someone just sent you the pickle .pkl file for you to work with.
What you can do is to re-create a valid Pandas dataframe as it was with the training/fitting. e.g.
# features here is just a list
df = pandas.DataFrame([features])
df.columns = ["provide", "the", "list", "of", "feature names", "in", "order"]
prediction = model.predict(df)I got the same error while using dataframes but by passing only values it is no more there
use
reg = reg.predict( x[['data']].values , y)It is showing error because our dataframe has feature names but we should fit the data as 2d array(or matrix) with values for training or testing the dataset.
Here is the image of the same thing mentioned above image of jupytr notebook code
I would suggest also adding y.values to regression.fit() method, because if we are
In my example, I have a csv with three data points like so, Car,Model,Volume,Weight,CO2 Toyoty,Aygo,1000,790,99 Mitsubishi,Space Star,1200,1160,95
Based on an example, I am trying to find out how much CO2 a car produces based on its volume of engine and weight. So,
X = df[['Weight', 'Volume']]
y = df['CO2']
#python
regression = linear_model.LinearRegression()
# regression.fit(X, y) this gave some errors
regression.fit(X.values, y.values) # This did not give any errorsThe reason being grabbing the values will ignore the first line of csv data, i.e the title.
