Minor fixes reloading hyperlogger without stats logging

README.md

@@ -81,7 +81,7 @@ To cite this repository:
 }
 ```
 
-Much of the `mle-toolbox` design has been inspired by discussions with Jonathan Frankle and Nandan Rao about the quest for empirically sound and supported claims.
+Much of the `mle-toolbox` design has been inspired by discussions with [Jonathan Frankle](http://www.jfrankle.com/) and [Nandan Rao](https://twitter.com/nandanrao) about the quest for empirically sound and supported claims in Machine Learning.
 
 ## Notes, Development & Questions :question:
 

mle_toolbox/experiment/experiment.py

@@ -103,7 +103,7 @@ def run(self):
             job_id = self.schedule_remote()
             if self.job_status == 1:
                 self.logger.info(f"Job ID: {job_id} - Remote job scheduled" \
-                                  " - {self.config_filename}")
+                                 f" - {self.config_filename}")
             else:
                 self.logger.info(f"Job ID: {job_id} - Error when scheduling " \
                                  f"remote job - {self.config_filename}")
@@ -125,7 +125,7 @@ def run(self):
             status_out = self.monitor_local(proc)
             if status_out == 0:
                 self.logger.info("PID: {proc.pid} - Local job successfully " \
-                                 "completed - { self.config_filename}")
+                                 f"completed - { self.config_filename}")
             else:
                 self.logger.info(f"PID: {proc.pid} - Error when running local " \
                                  f"job - {self.config_filename}")
@@ -217,7 +217,8 @@ def cluster_available(self):
         else:
             self.run_on_sge_cluster = 0
             self.run_on_slurm_cluster = 0
-        return on_sge_cluster or on_sge_head or on_slurm_head or on_slurm_cluster
+        return (on_sge_cluster or on_sge_head or
+                on_slurm_head or on_slurm_cluster)
 
     def generate_cmd_line_args(self,
                                cmd_line_input: Union[None, dict]=None) -> str:

mle_toolbox/hyperopt/hyperlogger.py

@@ -17,7 +17,7 @@ def __init__(self, hyperlog_fname: str, max_target: bool=True,
         self.max_target = max_target  # Whether we want to max target (reward)
         self.eval_metrics = eval_metrics  # Var names to compare across runs
         self.verbose = verbose
-        self.no_results_logging = no_results_logging  # Option to not log metrics
+        self.no_results_logging = no_results_logging  # Want to not log metrics?
 
         # Instantiate the meta-logger
         self.logger = logging.getLogger(__name__)
@@ -90,7 +90,7 @@ def update_log(self, params, meta_eval_log,
 
     def print_log_state(self):
         """ Log currently best param config for each metric. """
-        if self.iter_id > 0:
+        if self.iter_id > 0 and not self.no_results_logging:
             for i, m in enumerate(self.best_per_metric.keys()):
                 print_framed(m, frame_str="-")
                 self.logger.info(
@@ -123,7 +123,8 @@ def reload_log(self):
             self.all_run_ids = []
 
         # Get best performing params for each eval metric
-        self.best_per_metric = self.get_best_performances(self.eval_metrics)
+        if not self.no_results_logging:
+            self.best_per_metric = self.get_best_performances(self.eval_metrics)
 
     def get_best_performances(self, eval_metrics):
         """ Get best performing hyperparam configuration up to current iter """

mle_toolbox/launch/prepare_experiment.py

@@ -23,7 +23,7 @@ def welcome_to_mle_toolbox(verbose=False):
     print(85*"=")
     time_t = datetime.now().strftime("%m/%d/%Y %I:%M:%S %p")
     print(time_t, f"Thx for using MLE-Toolbox {__version__}"
-                  f" Locally, on SGE or Slurm or Clusters.")
+                  f" Locally, on Clusters or the Cloud.")
     if verbose:
         print(time_t, "It implements the following experiment types:")
         print("  - single-experiment: Run a single configuration experiment.")

mle_toolbox/utils/mle_experiment.py

@@ -3,13 +3,15 @@
                               get_extra_cmd_line_input,
                               set_random_seeds)
 from .mle_logger import MLE_Logger
+from .helpers import print_framed
 
 
 class MLExperiment(object):
     def __init__(self,
                  config_fname: str="configs/base_config.json",
                  auto_setup: bool=True,
-                 create_jax_prng: bool=False):
+                 create_jax_prng: bool=False,
+                 default_seed: int=0):
         ''' Load the job configs for the MLE experiment. '''
         # Load the different configurations for the experiment.
         train_config, net_config, log_config, extra_args = get_configs_ready(
@@ -23,13 +25,20 @@ def __init__(self,
         self.log_config = log_config
         self.extra_config = extra_config
         self.create_jax_prng = create_jax_prng
+        self.default_seed = default_seed
 
         # Make initial setup optional so that configs can be modified ad-hoc
         if auto_setup:
             self.setup()
 
     def setup(self):
         ''' Set the random seed & initialize the logger. '''
+        # If no seed is provided in train_config - set it to default
+        if "seed_id" not in self.train_config.keys():
+            self.train_config.seed_id = self.default_seed
+            print_framed(f"!!!WARNING!!!: No seed provided - set to default "
+                         f"{self.default_seed}.")
+
         # Set the random seeds for all random number generation
         if self.create_jax_prng:
             # Return JAX random number generating key