Set value of first column from input

[ninamwa]

Hi.
I was wondering if is possible to set the values of the first columns from user input or other data?

Ex: I want to generate 10 new rows of data, but i do not want the age of each person (which is the first column) to be randomly sampled from the observed data, i want it to be set based on input from the user :-)

[ganevgv]

Hi @ninamwa,

Thanks for your question. Unfortunately, the current version doesn't support this functionality. However, I managed to make some quick and dirty changes that will do the job.

1. When installing the package from source - please make sure that you install with develop. I.e. run python setup.py develop instead of python setup.py install

2. Make the following changes to the synthpop/synthpop.py file. I.e. delete all the lines in red and add all the lines in green.

diff --git a/synthpop/synthpop.py b/synthpop/synthpop.py
index 0b36c0e..e72b56a 100644
--- a/synthpop/synthpop.py
+++ b/synthpop/synthpop.py
@@ -81,11 +81,11 @@ class Synthpop:
             # save the method
             self.saved_methods[col] = col_method

-    def generate(self, k=None):
-        self.k = k
+    def generate(self, condition_df):
+        self.condition_df = condition_df

         # check generate
-        self.validator.check_generate()
+        # self.validator.check_generate()
         # generate
         synth_df = self._generate()
         # postprocess
@@ -94,9 +94,12 @@ class Synthpop:
         return processed_synth_df

     def _generate(self):
-        synth_df = pd.DataFrame(data=np.zeros([self.k, len(self.visit_sequence)]), columns=self.visit_sequence.index)
+        synth_df = pd.DataFrame(data=np.zeros([len(self.condition_df), len(self.visit_sequence)]), columns=self.visit_sequence.index)
+        synth_df[list(self.condition_df.columns)] = self.condition_df

         for col, visit_step in self.visit_sequence.sort_values().iteritems():
+            if col in list(self.condition_df.columns):
+                continue
             print('generate_{}'.format(col))

             # reload the method

3. Here's an example if you want to generate synth data with first column containing [25, 80, 20, 50, 47] ages.

import pandas as pd

from synthpop import Synthpop
from datasets.adult import df, dtypes


spop = Synthpop()
spop.fit(df, dtypes)

condition_df = pd.DataFrame({"age": [25, 80, 20, 50, 47]})
synth_df = spop.generate(condition_df)
print(synth_df.head())

Hope that helps.

[ninamwa]

Thank you very much, this was exactly what i needed :-)

[ninamwa]

Hi again.
I also have a question about the implementation.
Why have you chosen to use the apply method for the decision trees when predicting new data, instead of simply using the built-in predict method?

I am guessing it has something to do with the ability to use random.choice for the possibilities in each leaf, instead of getting the same result each time, but i was not sure :)

[ganevgv]

That's exactly right. My goal was to replicate the methodology from the R package and this is the authors' approach.