Update nb & fix coordinate update

README.md

@@ -3,7 +3,7 @@
 [![PyPI version](https://badge.fury.io/py/mle-hyperopt.svg)](https://badge.fury.io/py/mle-hyperopt)
 [![Code style: black](https://img.shields.io/badge/code%20style-black-000000.svg)](https://github.com/psf/black)
 [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/RobertTLange/mle-hyperopt/blob/main/examples/getting_started.ipynb)
-<a href="docs/logo_transparent.png_2"><img src="docs/logo_transparent.png" width="200" align="right" /></a>
+<a href="https://github.com/RobertTLange/mle-hyperopt/blob/main/docs/logo_transparent.png?raw=true"><img src="https://github.com/RobertTLange/mle-hyperopt/blob/main/docs/logo_transparent.png?raw=true" width="200" align="right" /></a>
 
 The `mle-hyperopt` package provides a simple and intuitive API for hyperparameter optimization of your Machine Learning Experiment (MLE) pipeline. It supports real, integer & categorical search variables and single- or multi-objective optimization.
 
@@ -131,10 +131,10 @@ strategy.print_ranking(top_k=3)
 # Refine the search space after 5 & 10 iterations based on top 2 configurations
 strategy = RandomSearch(real={"lrate": {"begin": 0.1,
                                         "end": 0.5,
-                                        "prior": "uniform"}},
+                                        "prior": "log-uniform"}},
                         integer={"batch_size": {"begin": 1,
                                                 "end": 5,
-                                                "prior": "log-uniform"}},
+                                                "prior": "uniform"}},
                         categorical={"arch": ["mlp", "cnn"]},
                         search_config={"refine_after": [5, 10],
                                        "refine_top_k": 2})
@@ -144,15 +144,8 @@ strategy.tell(...)
 strategy.refine(top_k=2)
 ```
 
-Note the search space refinement is only implemented for random, SMBO and nevergrad-based search strategies.
+Note that the search space refinement is only implemented for random, SMBO and nevergrad-based search strategies.
 
 ## Development & Milestones for Next Release
 
-You can run the test suite via `python -m pytest -vv tests/`. If you find a bug or are missing your favourite feature, feel free to contact me [@RobertTLange](https://twitter.com/RobertTLange) or create an issue :hugs:. Here are some features I want to implement for the next release:
-
-- [ ] Add text to notebook for what is implemented
-- [ ] Update Readme text
-- [ ] Update mle-toolbox webpage intro
-- [ ] Release and make sure installation works
-- [ ] Draft tweet for release
-- [ ] Synergies with mle-logging
+You can run the test suite via `python -m pytest -vv tests/`. If you find a bug or are missing your favourite feature, feel free to contact me [@RobertTLange](https://twitter.com/RobertTLange) or create an issue :hugs:.

mle_hyperopt/_version.py

@@ -1 +1 @@
-__version__ = "0.0.2"
+__version__ = "0.0.3"

mle_hyperopt/search.py

@@ -5,6 +5,7 @@
 from mle_hyperopt.comms import welcome_message, update_message, ranking_message
 import matplotlib.pyplot as plt
 import seaborn as sns
+from rich.console import Console
 
 sns.set(
     context="poster",
@@ -129,7 +130,7 @@ def tell(
                     del proposal_clean[k]
 
             if proposal_clean in self.all_evaluated_params:
-                print(f"{batch_proposals[i]} was previously evaluated.")
+                Console().log(f"{batch_proposals[i]} was previously evaluated.")
             else:
                 self.log.append(
                     {
@@ -170,7 +171,9 @@ def save(self, save_path: str = "search_log.json", verbose: bool = False):
         """Store the state of the optimizer (parameters, values) as .pkl."""
         save_json(self.log, save_path)
         if verbose:
-            print(f"Stored {self.eval_counter} search iterations --> {save_path}.")
+            Console().log(
+                f"Stored {self.eval_counter} search iterations --> {save_path}."
+            )
 
     def load(
         self,
@@ -190,7 +193,7 @@ def load(
                 self.tell([iter["params"]], [iter["objective"]], True)
 
         if reload_path is not None or reload_list is not None:
-            print(
+            Console().log(
                 f"Reloaded {self.eval_counter - prev_evals}"
                 " previous search iterations."
             )
@@ -254,7 +257,8 @@ def plot_best(self):
             timeseries = np.maximum.accumulate(objective_evals)
 
         fig, ax = plt.subplots()
-        ax.plot(timeseries)
+        # Use rich logging!!!
+        ax.plot(np.arange(1, len(timeseries) + 1), timeseries)
         ax.spines["top"].set_visible(False)
         ax.spines["right"].set_visible(False)
         ax.set_title("Best Objective Value")

mle_hyperopt/strategies/coordinate.py

@@ -2,6 +2,7 @@
 from ..search import HyperOpt
 from ..spaces import GridSpace
 import numpy as np
+from rich.console import Console
 
 
 class CoordinateSearch(HyperOpt):
@@ -32,27 +33,22 @@ def __init__(
             verbose,
         )
         self.evals_per_coord = [0]
+        var_counter = 0
         for k in self.search_config["order"]:
             if self.real is not None:
                 if k in self.real.keys():
-                    self.evals_per_coord.append(self.real[k]["bins"])
+                    self.evals_per_coord.append(self.real[k]["bins"] + var_counter)
+                    var_counter += 1
 
             if self.integer is not None:
                 if k in self.integer.keys():
-                    range_int = (
-                        np.linspace(
-                            int(self.integer[k]["begin"]),
-                            int(self.integer[k]["end"]),
-                            int(self.integer[k]["bins"]),
-                        )
-                        .astype(int)
-                        .tolist()
-                    )
-                    self.evals_per_coord.append(len(range_int))
+                    self.evals_per_coord.append(self.integer[k]["bins"] + var_counter)
+                    var_counter += 1
 
             if self.categorical is not None:
                 for k in self.categorical.keys():
-                    self.evals_per_coord.append(len(self.categorical[k]))
+                    self.evals_per_coord.append(len(self.categorical[k]) + var_counter)
+                    var_counter += 1
         self.range_per_coord = np.cumsum(self.evals_per_coord)
 
         # Sequentially set-up different grid spaces - initialize 1st one
@@ -95,7 +91,7 @@ def tell_search(self, batch_proposals: list, perf_measures: list):
         self.grid_var_counter = (
             self.eval_counter - self.range_per_coord[self.var_counter]
         )
-        if self.grid_var_counter == len(self.space):
+        if self.grid_var_counter >= len(self.space) - self.var_counter:
             self.var_counter += 1
             if self.var_counter < len(self.search_config["order"]):
                 self.construct_active_space()
@@ -107,10 +103,15 @@ def construct_active_space(self):
         if self.eval_counter > 0:
             idx, config, eval = self.get_best()
             for k, v in config.items():
-                self.search_config["defaults"][k] = v
+                if k == self.search_config["order"][self.var_counter - 1]:
+                    self.search_config["defaults"][k] = v
+                    if self.verbose:
+                        Console().log(f"Fixed `{k}` hyperparameter to {v}.")
 
         # Increase active variable counter and reset grid counter
         self.active_var = self.search_config["order"][self.var_counter]
+        if self.verbose:
+            Console().log(f"New active variable `{self.active_var}`.")
 
         # Create new grid search space - if fixed: Create categorical
         # Note: Only one variable is 'active' at every time

mle_hyperopt/strategies/nevergrad.py

@@ -43,6 +43,11 @@ def __init__(
         if self.verbose:
             self.print_hello()
 
+    @property
+    def optimizers(self):
+        """ Returns list of available nevergrad optimizers. """
+        return sorted(ng.optimizers.registry.keys())
+
     def init_optimizer(self):
         """Initialize the surrogate model/hyperparam config proposer."""
         assert self.search_config["optimizer"] in list(dict(ng.optimizers.registry).keys())