kaknikhil commented on a change in pull request #443: DL: Add training for
multiple models
URL: https://github.com/apache/madlib/pull/443#discussion_r325810826
##########
File path: src/ports/postgres/modules/deep_learning/madlib_keras.py_in
##########
@@ -387,70 +396,91 @@ def should_compute_metrics_this_iter(curr_iter,
metrics_compute_frequency,
return (curr_iter)%metrics_compute_frequency == 0 or \
curr_iter == num_iterations
+def init_model(model_architecture, compile_params):
+ """
+ Should only be called at the first row of first iteration.
+ """
+ segment_model = model_from_json(model_architecture)
+ compile_model(segment_model, compile_params)
+ return segment_model
+
+def update_model(segment_model, prev_serialized_weights):
+ """
+ Happens at first row of each iteration.
+ """
+ model_shapes = get_model_shapes(segment_model)
+ model_weights = madlib_keras_serializer.deserialize_as_nd_weights(
+ prev_serialized_weights, model_shapes)
+ segment_model.set_weights(model_weights)
+
def fit_transition(state, dependent_var, independent_var, dependent_var_shape,
independent_var_shape, model_architecture,
compile_params, fit_params, current_seg_id, seg_ids,
images_per_seg, gpus_per_host, segments_per_host,
- prev_serialized_weights, **kwargs):
+ prev_serialized_weights, is_final_iteration=True,
+ is_multiple_model=False, **kwargs):
Review comment:
I think we should add a docstring for this function and explain how this
function is used for `madlib_keras_fit` vs `madlib_keras_fit_multiple`. We can
also explain how the tf sessions are set up for fit/eval vs
fit_multiple/eval_multiple.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
