Added rank correlation plots based on BOHB from HpBandster

.gitignore

@@ -23,3 +23,7 @@ dask-worker-space/
 dehb/examples/*/results
 dehb/examples/*/*/results
 .ipynb_checkpoints/
+temp/
+examples/*/results
+examples/*/*/results
+examples/data

dehb/utils/__init__.py

@@ -1 +1,2 @@
-from .bracket_manager import SHBracketManager
+from .bracket_manager import SHBracketManager
+from .analysis import *

dehb/utils/analysis.py

@@ -0,0 +1,125 @@
+import scipy.stats as sps
+import matplotlib.pyplot as plt
+import pickle
+import numpy as np
+
+
+def load_pickle(filename):
+
+    with open(filename, "rb") as f:
+        var = pickle.load(f)
+
+    return var
+
+
+def correlation_across_budgets(filename=None, show=False):
+    '''
+    Based on code from here: 
+    https://automl.github.io/HpBandSter/build/html/_modules/hpbandster/visualization.html#correlation_across_budgets
+
+
+    Parameters
+    ----------
+    filename : str
+        If filename is None, it uses the latest modified .pkl file in the sub-directory temp of the current working directory
+    show : bool
+        Whether to show resulting plot or not
+
+    Returns
+    -------
+    tuple of (matplotlib.figure.Figure, matplotlib.axes._subplots.AxesSubplot)
+
+
+    '''
+
+    if filename is None:
+        import os
+        os.chdir('temp')
+        files = os.listdir()
+        files.sort(key=os.path.getmtime)
+        for file in files[::-1]:
+            if file[-4:] == '.pkl':
+                filename = file[:-4]
+                break
+        os.chdir('..')
+
+        print('No filename provided. Loading the latest modified .pkl file in sub-directory '
+                'temp with name: ' + filename + '.pkl.')
+
+    complete_trajectory = load_pickle("./temp/" + filename + ".pkl")
+    print("Loaded trajectory from pickle:", complete_trajectory)
+
+    stats_dict = {}
+    for r in complete_trajectory:
+        # Check if config is already in our stats dict
+        if str(r[0]) not in stats_dict:
+            stats_dict[str(r[0])] = []
+
+        stats_dict[str(r[0])].append({"fitness": r[1], "cost": r[2], "budget": r[3]})
+
+
+    budgets = list(set([r[3] for r in complete_trajectory]))
+    budgets.sort()
+
+    import itertools
+
+    loss_pairs = {}
+    for b in budgets[:-1]:
+        loss_pairs[b] = {}
+
+    for b1, b2 in itertools.combinations(budgets, 2):
+        loss_pairs[b1][b2]= []
+    print("loss_pairs:", loss_pairs)
+
+    for config in stats_dict:
+        print(stats_dict[config])
+        if len(stats_dict[config]) < 2: continue
+
+        for r1, r2 in itertools.combinations(stats_dict[config], 2):
+            if not np.isfinite(r1["fitness"]) or not np.isfinite(r2["fitness"]): continue
+            loss_pairs[float(r1["budget"])][float(r2["budget"])].append((r1["fitness"], r2["fitness"]))
+
+    print("budgets:", budgets)
+    print("loss_pairs:", loss_pairs)
+
+
+    rhos = np.eye(len(budgets)-1)
+    rhos.fill(np.nan)
+
+    ps = np.eye(len(budgets)-1)
+    ps.fill(np.nan)
+
+    for i in range(len(budgets)-1):
+        for j in range(i+1,len(budgets)):
+            spr = sps.spearmanr(loss_pairs[budgets[i]][budgets[j]])
+            rhos[i][j-1] = spr.correlation
+            ps[i][j-1] = spr.pvalue
+
+
+    fig, ax = plt.subplots()
+
+    cax = ax.matshow(rhos, vmin=-1, vmax=1)
+    fig.colorbar(cax)
+
+
+    ax.set_yticks( range(len(budgets)-1))
+    ax.set_yticklabels(budgets[:-1],)
+
+    ax.set_xticks( range(len(budgets)-1))
+    ax.set_xticklabels(budgets[1:],)
+
+    ax.set_title('Rank correlation of the loss across the budgets')
+
+    for i in range(len(budgets)-1):
+        for j in range(i+1,len(budgets)):
+            plt.text(j-1,i, r'$\rho_{spearman}= %f$'%rhos[i][j-1] + '\n' + r'$p = %f$'%ps[i][j-1] + 
+                        '\n' + r'$n = %i$'%len(loss_pairs[budgets[i]][budgets[j]]),
+                        horizontalalignment='center', verticalalignment='center')
+
+
+    if show:
+        plt.show()
+
+    print(type(fig), type(ax))
+    return(fig, ax)
+