Github user njayaram2 commented on a diff in the pull request:
https://github.com/apache/madlib/pull/179#discussion_r137672182
--- Diff: src/ports/postgres/modules/convex/mlp_igd.py_in ---
@@ -158,175 +236,257 @@ def mlp(schema_madlib, source_table, output_table,
independent_varname,
# See design doc for more info
span = math.sqrt(6.0 / (fan_in + fan_out))
dim = (layer_sizes[i] + 1) * layer_sizes[i + 1]
- rand = plpy.execute("""SELECT
array_agg({span}*(random()-0.5))
+ rand = plpy.execute("""SELECT
array_agg({span}*2*(random()-0.5))
--- End diff --
No, there does not seem to be any particular reason. I think @cooper-sloan
had modified it.
It just changes the initial random weights for the coefficients generated,
and does not seem to affect the algorithm's convergence all that much.
---
