Reputation: 1219
I have a df column that has values ranging from -5 to 10. I want to change values <= -1 to negative
, all 0 values to neutral
, and all values >= 1 to positive
. The code below, however, produces the following error for 'negative'.
# Function to change values to labels
test.loc[test['sentiment_score'] > 0, 'sentiment_score'] = 'positive'
test.loc[test['sentiment_score'] == 0, 'sentiment_score'] = 'neutral'
test.loc[test['sentiment_score'] < 0, 'sentiment_score'] = 'negative'
Data: Data After Code:
Index Sentiment Index Sentiment
0 2 0 positive
1 0 1 neutral
2 -3 2 -3
3 4 3 positive
4 -1 4 -1
... ...
k 5 k positive
File "pandas_libs\ops.pyx", line 98, in pandas._libs.ops.scalar_compare TypeError: '<=' not supported between instances of 'str' and 'int
I assume that this has something to do with the function seeing negative numbers as string rather than float/int, however I've tried the following code to correct this error and it changes nothing. Any help would be appreciated.
test['sentiment_score'] = test['sentiment_score'].astype(float)
test['sentiment_score'] = test['sentiment_score'].apply(pd.as_numeric)
Upvotes: 3
Views: 14219
Reputation: 183
Another alternative is to define a custom function:
def transform_sentiment(x):
if x < 0:
return 'Negative'
elif x == 0:
return 'Neutral'
else:
return 'Positive'
df['Sentiment_new'] = df['Sentiment'].apply(lambda x: transform_sentiment(x))
Upvotes: 0
Reputation: 403278
As roganjosh pointed out, you're doing your replacement in 3 steps - this is causing a problem because after step 1, you end up with a column of mixed dtypes, so subsequent equality checks start to fail.
You can either assign to a new column, or use numpy.select
.
condlist = [
test['sentiment_score'] > 0,
test['sentiment_score'] < 0
]
choicelist = ['pos', 'neg']
test['sentiment_score'] = np.select(
condlist, choicelist, default='neutral')
Upvotes: 6