How to pivot dataframe into ML format
Question:
My head is spinning trying to figure out if I have to use pivot_table, melt, or some other function.
I have a DF that looks like this:
month day week_day classname_en origin destination
0 1 7 2 1 2 5
1 1 2 6 2 1 167
2 2 1 5 1 2 54
3 2 2 6 4 1 6
4 1 2 6 5 6 1
But I want to turn it into something like:
month_1 month_2 ...classname_en_1 classname_en_2 ... origin_1 origin_2 ...destination_1
0 1 0 1 0 0 1 0
1 1 0 0 1 1 0 0
2 0 1 1 0 0 1 0
3 0 1 0 0 1 0 0
4 1 0 0 0 0 0 1
Basically, turn all values into columns and then have binary rows 1 – if the column is present, 0 if none.
IDK if it is at all possible to do with like a single function or not, but would appreciate all and any help!
Answers:
Use pd.get_dummies
:
out = pd.get_dummies(df, columns=df.columns)
print(out)
# Output
month_1 month_2 day_1 day_2 day_7 week_day_2 week_day_5 ... origin_2 origin_6 destination_1 destination_5 destination_6 destination_54 destination_167
0 1 0 0 0 1 1 0 ... 1 0 0 1 0 0 0
1 1 0 0 1 0 0 0 ... 0 0 0 0 0 0 1
2 0 1 1 0 0 0 1 ... 1 0 0 0 0 1 0
3 0 1 0 1 0 0 0 ... 0 0 0 0 1 0 0
4 1 0 0 1 0 0 0 ... 0 1 1 0 0 0 0
[5 rows x 20 columns]
To expand @Corraliens answer
It is indeed a way to do it, but since you write for ML purposes, you might introduce a bug.
With the code above you get a matrix with 20 features. Now, say you want to predict on some data which suddenly have a month more than your training data, then your matrix on your prediction data would have 21 features, thus you cannot parse that into your fitted model.
To overcome this you can use one-hot-encoding from Sklearn. It’ll make sure that you always have the same amount of features on "new data" as your training data.
import pandas as pd
df_train = pd.DataFrame({"color":["red","blue"],"age":[10,15]})
pd.get_dummies(df_train)
# output
age color_blue color_red
0 10 0 1
1 15 1 0
df_new = pd.DataFrame({"color":["red","blue","green"],"age":[10,15,20]})
pd.get_dummies(df_new)
#output
age color_blue color_green color_red
0 10 0 0 1
1 15 1 0 0
2 20 0 1 0
and as you can see, the order of the color-binary representation has also changed.
If we on the other hand use OneHotEncoder
you can ommit all those issues
from sklearn.preprocessing import OneHotEncoder
df_train = pd.DataFrame({"color":["red","blue"],"age":[10,15]})
ohe = OneHotEncoder(handle_unknown="ignore")
color_ohe_transformed= ohe.fit_transform(df_train[["color"]]) #creates sparse matrix
ohe_features = ohe.get_feature_names_out() # [color_blue, color_red]
pd.DataFrame(color_ohe_transformed.todense(),columns = ohe_features, dtype=int)
# output
color_blue color_red
0 0 1
1 1 0
# now transform new data
df_new = pd.DataFrame({"color":["red","blue","green"],"age":[10,15,20]})
new_data_ohe_transformed = ohe.transform(df_new[["color"]])
pd.DataFrame(new_data_ohe_transformed .todense(),columns = ohe_features, dtype=int)
#output
color_blue color_red
0 0 1
1 1 0
2 0 0
note in the last row that both blue
and red
are both zeros since it has color= "green"
which was not present in the training data.
Note the todense()
function is only used here to illustrate how it works. Ususally you would like to keep it a sparse matrix and use e.g scipy.sparse.hstack
to append your other features such as age
to it.
You can use get_dummies function of pandas for convert row to column based on data.
For that your code will be:
import pandas as pd
df = pd.DataFrame({
'month': [1, 1, 2, 2, 1],
'day': [7, 2, 1, 2, 2],
'week_day': [2, 6, 5, 6, 6],
'classname_en': [1, 2, 1, 4, 5],
'origin': [2, 1, 2, 1, 6],
'destination': [5, 167, 54, 6, 1]
})
response = pd.get_dummies(df, columns=df.columns)
print(response)
My head is spinning trying to figure out if I have to use pivot_table, melt, or some other function.
I have a DF that looks like this:
month day week_day classname_en origin destination 0 1 7 2 1 2 5 1 1 2 6 2 1 167 2 2 1 5 1 2 54 3 2 2 6 4 1 6 4 1 2 6 5 6 1
But I want to turn it into something like:
month_1 month_2 ...classname_en_1 classname_en_2 ... origin_1 origin_2 ...destination_1 0 1 0 1 0 0 1 0 1 1 0 0 1 1 0 0 2 0 1 1 0 0 1 0 3 0 1 0 0 1 0 0 4 1 0 0 0 0 0 1
Basically, turn all values into columns and then have binary rows 1 – if the column is present, 0 if none.
IDK if it is at all possible to do with like a single function or not, but would appreciate all and any help!
Use pd.get_dummies
:
out = pd.get_dummies(df, columns=df.columns)
print(out)
# Output
month_1 month_2 day_1 day_2 day_7 week_day_2 week_day_5 ... origin_2 origin_6 destination_1 destination_5 destination_6 destination_54 destination_167
0 1 0 0 0 1 1 0 ... 1 0 0 1 0 0 0
1 1 0 0 1 0 0 0 ... 0 0 0 0 0 0 1
2 0 1 1 0 0 0 1 ... 1 0 0 0 0 1 0
3 0 1 0 1 0 0 0 ... 0 0 0 0 1 0 0
4 1 0 0 1 0 0 0 ... 0 1 1 0 0 0 0
[5 rows x 20 columns]
To expand @Corraliens answer
It is indeed a way to do it, but since you write for ML purposes, you might introduce a bug.
With the code above you get a matrix with 20 features. Now, say you want to predict on some data which suddenly have a month more than your training data, then your matrix on your prediction data would have 21 features, thus you cannot parse that into your fitted model.
To overcome this you can use one-hot-encoding from Sklearn. It’ll make sure that you always have the same amount of features on "new data" as your training data.
import pandas as pd
df_train = pd.DataFrame({"color":["red","blue"],"age":[10,15]})
pd.get_dummies(df_train)
# output
age color_blue color_red
0 10 0 1
1 15 1 0
df_new = pd.DataFrame({"color":["red","blue","green"],"age":[10,15,20]})
pd.get_dummies(df_new)
#output
age color_blue color_green color_red
0 10 0 0 1
1 15 1 0 0
2 20 0 1 0
and as you can see, the order of the color-binary representation has also changed.
If we on the other hand use OneHotEncoder
you can ommit all those issues
from sklearn.preprocessing import OneHotEncoder
df_train = pd.DataFrame({"color":["red","blue"],"age":[10,15]})
ohe = OneHotEncoder(handle_unknown="ignore")
color_ohe_transformed= ohe.fit_transform(df_train[["color"]]) #creates sparse matrix
ohe_features = ohe.get_feature_names_out() # [color_blue, color_red]
pd.DataFrame(color_ohe_transformed.todense(),columns = ohe_features, dtype=int)
# output
color_blue color_red
0 0 1
1 1 0
# now transform new data
df_new = pd.DataFrame({"color":["red","blue","green"],"age":[10,15,20]})
new_data_ohe_transformed = ohe.transform(df_new[["color"]])
pd.DataFrame(new_data_ohe_transformed .todense(),columns = ohe_features, dtype=int)
#output
color_blue color_red
0 0 1
1 1 0
2 0 0
note in the last row that both blue
and red
are both zeros since it has color= "green"
which was not present in the training data.
Note the todense()
function is only used here to illustrate how it works. Ususally you would like to keep it a sparse matrix and use e.g scipy.sparse.hstack
to append your other features such as age
to it.
You can use get_dummies function of pandas for convert row to column based on data.
For that your code will be:
import pandas as pd
df = pd.DataFrame({
'month': [1, 1, 2, 2, 1],
'day': [7, 2, 1, 2, 2],
'week_day': [2, 6, 5, 6, 6],
'classname_en': [1, 2, 1, 4, 5],
'origin': [2, 1, 2, 1, 6],
'destination': [5, 167, 54, 6, 1]
})
response = pd.get_dummies(df, columns=df.columns)
print(response)