Update QA_SVM.ipynb

51840204 · Amer Delilbasic · 3b28fe37 · 51840204
Commit 51840204 authored 3 years ago by Amer Delilbasic
--- a/experiments/QA_SVM/QA_SVM.ipynb
+++ b/experiments/QA_SVM/QA_SVM.ipynb
@@ -4,7 +4,7 @@
   "cell_type": "markdown",
   "metadata": {},
   "source": [
-    "# Classification with SVM on the D-Wave Advantage Qauntum Annealer"
+    "# Classification with SVM on the D-Wave Advantage Quantum Annealer"
   ]
  },
  {

 %% Cell type:markdown id: tags:
  
-# Classification with SVM on the D-Wave Advantage Qauntum Annealer
+# Classification with SVM on the D-Wave Advantage Quantum Annealer
  
 %% Cell type:markdown id: tags:
  
 #### This notebook performs the evaluation and visualization of the results returned from the quantum annealer. The results have to be obtained with another script (e.g. submit_experiments.sh).
  
 The classification map that is reported in the paper is the one obtained with the following parameters:
  
 - B = 3
 - K = 2
 - gamma = -1
 - xi = 5
 - E = 0
 - embedding 2
 - annealing time = 100
 - rel. chain strength = 0.5
 - Energy -45.506845443310596
  
 On test data
 - Overal accuracy 0.873516
 - F1 score 0.7344402173456618
  
 %% Cell type:markdown id: tags:
  
 ## Import Packages
  
 %% Cell type:code id: tags:
  
 ``` python
 import numpy as np
 import glob
 from utils import *
 import matplotlib.pyplot as plt
 from matplotlib import colors
 from sklearn import preprocessing
  
 from sklearn.metrics import accuracy_score
 from sklearn.metrics import f1_score
  
 from quantum_SVM import *   # QA SVM
 ```
  
 %% Cell type:markdown id: tags:
  
 ## Load the training and test sets
  
 %% Cell type:code id: tags:
  
 ``` python
 # Load training set
 key='_tiny'
 X_train=np.load('X_train_tile_4'+key+'.npy')
 Y_train=np.load('Y_train_tile_4'+key+'.npy')
  
 print('X_train',X_train.shape)
 print('Y_train',Y_train.shape)
  
 # Load test set
 X_test=np.load('X_test_tile_8_subregion.npy')
 Y_test=np.load('Y_test_tile_8_subregion.npy')
  
 print('X_test',X_test.shape)
 print('Y_test',Y_test.shape)
  
 # for training, the labels have to be -1,+1; i.e., replace 0 -> -1
 Y_train=np.where(Y_train==0,-1,Y_train)
  
 groundtruth=np.reshape(Y_test,(500,500))
 plt.title('Groundtruth')
 cmap = colors.ListedColormap(['black', 'red'])
 plt.rcParams["figure.figsize"] = (5,5)
 view=plt.imshow(groundtruth,cmap=cmap)
 ```
  
 %% Output
  
    X_train (50, 8)
    Y_train (50,)
    X_test (250000, 8)
    Y_test (250000,)
  

  
 %% Cell type:markdown id: tags:
  
 ## QSVM
  
 %% Cell type:markdown id: tags:
  
 ### Parameters
  
 %% Cell type:code id: tags:
  
 ``` python
 outputpath='output/run_calibtrain'+key
 maxalphas=20 # the 20 lowest-energy results returned by the quantum annealer are stored, but for the evaluation, we can consider less and compare
  
 # Parameters
 Bs=[2,3] #[2,3,5,10]                      Base
 Ks=[2] #[2,3]                             Number of qubits
 xis=[0,1,5] #[0,1,5]                      Strength to consider the constraint
 gammas=[-1] #[-1,0.125,0.25,0.5,1,2,4,8]  Kernel
 Es=[0,1,2] #[0,1,2]                       Exponent
 annealing_times=[1,10,100]
 chain_strengths=[0.2,0.5,1,2,5]
 embeddings=[0,1,2,3]
 ```
  
 %% Cell type:markdown id: tags:
  
 ### Evaluation
  
 %% Cell type:code id: tags:
  
 ``` python
 cmap = colors.ListedColormap(['black', 'red'])
 plt.title('Prediction')
 plt.rcParams["figure.figsize"] = (5,5)
 Y_train_bin=np.where(Y_train==-1,0,Y_train)
  
 for B in Bs:
    for K in Ks:
        for gamma in gammas:
            for xi in xis:
                for E in Es:
                    dirs=glob.glob(outputpath+f'_B={B}_K={K}_xi={xi}_E={E}_gamma={gamma}/result_couplers=*')
                    if not dirs:
                        dirs=glob.glob(outputpath+f'_B={B}_K={K}_xi={xi}_E={E}_gamma={float(gamma)}/result_couplers=*')
                    path=dirs[0]+'/'
                    f = open(path+f'collected_data_all_embeddings_maxalphas{maxalphas}.txt',"w")
                    f.write("#rcs \tt_a \t trainacc\t trainF1score\t testacc\t testF1score\t average energy(train)\n")
                    for emb in embeddings:
                        for c in chain_strengths:
                            for t in annealing_times:
                                alphas=np.load(path+f'embedding{emb}_rcs{c}_ta{t}_alphas.npy')
                                if not maxalphas == 0 or maxalphas > len(alphas):
                                    alphas = alphas[0:maxalphas]
  
                                scores_train=predict(X_train,X_train,Y_train,alphas,path)
                                Y_predict_train=np.sign(scores_train)
                                Y_predict_train=np.where(Y_predict_train==-1,0,Y_predict_train)
                                Y_predict_train=np.where(Y_predict_train==1,1,Y_predict_train)
  
                                scores=predict(X_test,X_train,Y_train,alphas,path)
                                Y_predict=np.sign(scores)
                                Y_predict=np.where(Y_predict==-1,0,Y_predict)   # From -1 to 0
                                Y_predict=np.where(Y_predict==1,1,Y_predict)    # From -1 to 1
  
                                trainacc = accuracy_score(Y_train_bin[:], Y_predict_train)
                                trainF1score = f1_score(Y_train_bin[:], Y_predict_train)
                                testacc = accuracy_score(Y_test[:], Y_predict)
                                testF1score = f1_score(Y_test[:], Y_predict)
                                alphas_avg = np.mean(alphas,axis=0)
                                av_energy = compute_energy(alphas_avg,X_train,Y_train,gamma,xi)
  
                                f.write(f'{c:1.2f}\t {t:4}\t {trainacc:8.4f}\t{trainF1score:8.4f}\t{testacc:8.4f}\t{testF1score:8.4f}\t{av_energy:8.4f}')
                                f.write("\n")
  
                                #  Visualize the prediction only for reasonable solutions
                                if testacc > 0.75 and testF1score > 0.7:
                                    print(f'B = {B}, K = {K}, gamma = {gamma}, xi = {xi}, E = {E},\n  embedding {emb}, annealing time = {t}, rel. chain strength = {c}')
                                    #print('On train data:')
                                    #print ('Overal accuracy',trainacc)
                                    #print ('F1 score',trainF1score)
                                    print('Energy',av_energy)
                                    print('On test data')
                                    print ('Overal accuracy',testacc)
                                    print ('F1 score',testF1score)
                                    classification_map=np.reshape(Y_predict,(500,500))
                                    plt.imshow(classification_map, cmap=cmap)
                                    plt.clim(0, 1)
                                    plt.show()
  
                            f.write("\n")
                        f.write("\n")
                    f.close()
  
 ```
  
 %% Output
  
    B = 2, K = 2, gamma = -1, xi = 0, E = 0,
      embedding 0, annealing time = 100, rel. chain strength = 0.5
    Energy -63.576530966387956
    On test data
    Overal accuracy 0.86842
    F1 score 0.7326283619575554
  

  
    B = 2, K = 2, gamma = -1, xi = 0, E = 0,
      embedding 3, annealing time = 1, rel. chain strength = 0.5
    Energy -70.07246567078842
    On test data
    Overal accuracy 0.84796
    F1 score 0.7101128737034776
  

  
    B = 2, K = 2, gamma = -1, xi = 0, E = 0,
      embedding 3, annealing time = 100, rel. chain strength = 0.5
    Energy -64.84508239813442
    On test data
    Overal accuracy 0.877876
    F1 score 0.7158400268049105
  

  
    B = 2, K = 2, gamma = -1, xi = 0, E = 1,
      embedding 2, annealing time = 10, rel. chain strength = 0.2
    Energy -38.65519361568767
    On test data
    Overal accuracy 0.868012
    F1 score 0.7030427387348471
  

  
    B = 2, K = 2, gamma = -1, xi = 0, E = 1,
      embedding 2, annealing time = 100, rel. chain strength = 0.2
    Energy -39.17549258303925
    On test data
    Overal accuracy 0.889452
    F1 score 0.7190133899976616
  

  
    B = 2, K = 2, gamma = -1, xi = 0, E = 2,
      embedding 0, annealing time = 10, rel. chain strength = 0.2
    Energy -20.652235656654433
    On test data
    Overal accuracy 0.888164
    F1 score 0.7211878857986219
  

  
    B = 2, K = 2, gamma = -1, xi = 1, E = 0,
      embedding 0, annealing time = 100, rel. chain strength = 0.5
    Energy -38.660751305593045
    On test data
    Overal accuracy 0.845076
    F1 score 0.7018651230457775
  

  
    B = 2, K = 2, gamma = -1, xi = 1, E = 0,
      embedding 1, annealing time = 10, rel. chain strength = 0.5
    Energy -28.59312172758268
    On test data
    Overal accuracy 0.845788
    F1 score 0.7008867958197238
  

  
    B = 2, K = 2, gamma = -1, xi = 1, E = 1,
      embedding 1, annealing time = 100, rel. chain strength = 0.5
    Energy -24.946913849585172
    On test data
    Overal accuracy 0.879816
    F1 score 0.726556243174372
  

  
    B = 2, K = 2, gamma = -1, xi = 1, E = 1,
      embedding 1, annealing time = 1, rel. chain strength = 1
    Energy -17.077607919194183
    On test data
    Overal accuracy 0.859328
    F1 score 0.7156624947446718
  

  
    B = 2, K = 2, gamma = -1, xi = 1, E = 1,
      embedding 2, annealing time = 10, rel. chain strength = 0.5
    Energy -25.241573461584892
    On test data
    Overal accuracy 0.865208
    F1 score 0.7028761881249229
  

  
    B = 2, K = 2, gamma = -1, xi = 1, E = 1,
      embedding 2, annealing time = 100, rel. chain strength = 0.5
    Energy -24.475775335015406
    On test data
    Overal accuracy 0.887352
    F1 score 0.7155008687921768
  

  
    B = 2, K = 2, gamma = -1, xi = 1, E = 1,
      embedding 3, annealing time = 10, rel. chain strength = 0.5
    Energy -20.584173298714248
    On test data
    Overal accuracy 0.873544
    F1 score 0.7167051992042585
  

  
    B = 2, K = 2, gamma = -1, xi = 1, E = 2,
      embedding 2, annealing time = 100, rel. chain strength = 0.2
    Energy -15.571795807843738
    On test data
    Overal accuracy 0.857188
    F1 score 0.7195871916307344
  

  
    B = 2, K = 2, gamma = -1, xi = 5, E = 0,
      embedding 0, annealing time = 10, rel. chain strength = 0.5
    Energy -49.52057699067239
    On test data
    Overal accuracy 0.88016
    F1 score 0.728608438864431
  

  
    B = 2, K = 2, gamma = -1, xi = 5, E = 0,
      embedding 0, annealing time = 1, rel. chain strength = 1
    Energy -35.00944361040178
    On test data
    Overal accuracy 0.854612
    F1 score 0.7104908918572327
  

  
    B = 2, K = 2, gamma = -1, xi = 5, E = 0,
      embedding 1, annealing time = 1, rel. chain strength = 0.5
    Energy -51.22353270783232
    On test data
    Overal accuracy 0.862656
    F1 score 0.7112000807455506
  

  
    B = 2, K = 2, gamma = -1, xi = 5, E = 0,
      embedding 1, annealing time = 10, rel. chain strength = 0.5
    Energy -46.22601878543993
    On test data
    Overal accuracy 0.862024
    F1 score 0.717900487421898
  

  
    B = 3, K = 2, gamma = -1, xi = 0, E = 0,
      embedding 1, annealing time = 10, rel. chain strength = 2
    Energy -71.4961865237801
    On test data
    Overal accuracy 0.840008
    F1 score 0.7030674664449459
  

  
    B = 3, K = 2, gamma = -1, xi = 0, E = 0,
      embedding 1, annealing time = 100, rel. chain strength = 5
    Energy -57.971641450473626
    On test data
    Overal accuracy 0.852416
    F1 score 0.7174971670086675
  

  
    B = 3, K = 2, gamma = -1, xi = 0, E = 0,
      embedding 3, annealing time = 1, rel. chain strength = 2
    Energy -78.40885247496453
    On test data
    Overal accuracy 0.865312
    F1 score 0.7132980263269927
  

  
    B = 3, K = 2, gamma = -1, xi = 0, E = 2,
      embedding 1, annealing time = 100, rel. chain strength = 1
    Energy -11.503963942188145
    On test data
    Overal accuracy 0.882028
    F1 score 0.7207128720371967
  

  
    B = 3, K = 2, gamma = -1, xi = 0, E = 2,
      embedding 1, annealing time = 1, rel. chain strength = 2
    Energy -11.399774035125132
    On test data
    Overal accuracy 0.882368
    F1 score 0.7165275394729232
  

  
    B = 3, K = 2, gamma = -1, xi = 0, E = 2,
      embedding 2, annealing time = 10, rel. chain strength = 0.2
    Energy -12.269963970450435
    On test data
    Overal accuracy 0.884696
    F1 score 0.7036679139767261
  

  
    B = 3, K = 2, gamma = -1, xi = 0, E = 2,
      embedding 2, annealing time = 100, rel. chain strength = 0.2
    Energy -12.051694743974974
    On test data
    Overal accuracy 0.884484
    F1 score 0.7337531230696894
  

  
    B = 3, K = 2, gamma = -1, xi = 1, E = 0,
      embedding 1, annealing time = 100, rel. chain strength = 0.5
    Energy -17.912004495524627
    On test data
    Overal accuracy 0.846016
    F1 score 0.7009833620729832
  

  
    B = 3, K = 2, gamma = -1, xi = 1, E = 1,
      embedding 0, annealing time = 1, rel. chain strength = 0.2
    Energy -18.20281620772596
    On test data
    Overal accuracy 0.86164
    F1 score 0.7187393277064936
  

  
    B = 3, K = 2, gamma = -1, xi = 1, E = 1,
      embedding 1, annealing time = 10, rel. chain strength = 0.5
    Energy -20.175029096824993
    On test data
    Overal accuracy 0.864616
    F1 score 0.7201144482667372
  

  
    B = 3, K = 2, gamma = -1, xi = 1, E = 1,
      embedding 2, annealing time = 100, rel. chain strength = 0.5
    Energy -20.659998098494363
    On test data
    Overal accuracy 0.884052
    F1 score 0.7282681040543707
  

  
    B = 3, K = 2, gamma = -1, xi = 1, E = 2,
      embedding 2, annealing time = 10, rel. chain strength = 0.5
    Energy -8.640766055370284
    On test data
    Overal accuracy 0.846384
    F1 score 0.7060363435954746
  

  
    B = 3, K = 2, gamma = -1, xi = 1, E = 2,
      embedding 2, annealing time = 100, rel. chain strength = 0.5
    Energy -8.614799617154471
    On test data
    Overal accuracy 0.873596
    F1 score 0.7318745280377401
  

  
    B = 3, K = 2, gamma = -1, xi = 1, E = 2,
      embedding 2, annealing time = 10, rel. chain strength = 1
    Energy -7.426846642711299
    On test data
    Overal accuracy 0.85826
    F1 score 0.7184055563943832
  

  
    B = 3, K = 2, gamma = -1, xi = 5, E = 0,
      embedding 2, annealing time = 100, rel. chain strength = 0.5
    Energy -45.506845443310596
    On test data
    Overal accuracy 0.873516
    F1 score 0.7344402173456618
  

  
    B = 3, K = 2, gamma = -1, xi = 5, E = 1,
      embedding 0, annealing time = 100, rel. chain strength = 0.5
    Energy -23.736552843941126
    On test data
    Overal accuracy 0.884704
    F1 score 0.7175225401803215
  

  
    B = 3, K = 2, gamma = -1, xi = 5, E = 1,
      embedding 2, annealing time = 10, rel. chain strength = 0.5
    Energy -23.863110771134867
    On test data
    Overal accuracy 0.88446
    F1 score 0.7171880354432859
  

  
    B = 3, K = 2, gamma = -1, xi = 5, E = 1,
      embedding 3, annealing time = 10, rel. chain strength = 0.2
    Energy -23.79218276655323
    On test data
    Overal accuracy 0.878908
    F1 score 0.7264545627050033
  

  
 %% Cell type:code id: tags:
  
 ``` python
 ```