[GitHub] [madlib] orhankislal commented on a change in pull request #524: DL: TensorBoard Support

[GitBox <gi...@apache.org>]

orhankislal commented on a change in pull request #524:
URL: https://github.com/apache/madlib/pull/524#discussion_r585517706



##########
File path: src/ports/postgres/modules/deep_learning/madlib_keras_wrapper.py_in
##########
@@ -287,27 +298,50 @@ def parse_optimizer(compile_dict):
 
 
 # Parse the fit parameters into a dictionary.
-def parse_and_validate_fit_params(fit_param_str):
+def parse_and_validate_fit_params(fit_param_str, current_seg_id=0):
 
     if fit_param_str:
-        fit_params_dict = convert_string_of_args_to_dict(fit_param_str)
-
-        literal_eval_fit_params = ['batch_size','epochs','verbose',
+        fit_params_dict = convert_string_of_args_to_dict(fit_param_str, strip_quotes=False)
+        literal_eval_fit_params = ['batch_size','epochs','verbose', 'shuffle',
                                    'class_weight','initial_epoch','steps_per_epoch']
-        accepted_fit_params = literal_eval_fit_params + ['shuffle']
+        accepted_fit_params = literal_eval_fit_params + ['callbacks']
 
         fit_params_dict = validate_and_literal_eval_keys(fit_params_dict,
                                                          literal_eval_fit_params,
                                                          accepted_fit_params)
-        if 'shuffle' in fit_params_dict:
-            shuffle_value = fit_params_dict['shuffle']
-            if shuffle_value == 'True' or shuffle_value == 'False':
-                fit_params_dict['shuffle'] = bool(shuffle_value)
+
+        if 'callbacks' in fit_params_dict:
+            fit_params_dict['callbacks'] = parse_callbacks(fit_params_dict['callbacks'], current_seg_id)
 
         return fit_params_dict
     else:
         return {}
 
+# Parse the callback fit params and create the TensorBoard object in the dictionary
+def parse_callbacks(callbacks, current_seg_id=0):
+    callbacks = callbacks.strip("'")
+    if not is_superuser(current_user()):
+        plpy.error("Only a superuser may use callbacks.")
+    try:
+        tree = ast.parse(callbacks, mode='eval')
+        assert(type(tree.body) == ast.List)
+        assert(len(tree.body.elts) == 1)
+        assert(type(tree.body.elts[0]) == ast.Call)
+        assert(tree.body.elts[0].func.id == 'TensorBoard')
+        tb_params = tree.body.elts[0].keywords
+        tb_params_dict = { tb_params[i].arg : tb_params[i].value \
+                        for i in range(len(tb_params)) }
+    except:
+        plpy.error("Invalid callbacks fit param.  Currently, "
+                    "only TensorBoard callbacks are accepted.")
+
+    accepted_tb_params = [ 'log_dir', 'histogram_freq', 'batch_size', 'update_freq',

Review comment:
       I don't think we have to test for every keras parameter combination. The `log_dir` is tested because it is error-prone with arbitrary paths but a numeric value like batch_size either works or keras throws an appropriate error. 




----------------------------------------------------------------
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:
users@infra.apache.org