Reproduction of Machine learning of high dimensional data on a noisy quantum processor.

python main.py [-h] [--run RUN] [--n_qubits N_QUBITS] [--n_samples N_SAMPLES] [--qgamma QGAMMA] [--qp QP] 
               [--rgamma RGAMMA] [--cv CV]

--run all to run all experiments. The program will go through the following steps：

1. Download PLAsTiCC data if no data available.

   The preset data urls may expire. Find data in kaggle.

2. Run process.py if no processed feature data to use.

3. Sample data with size n_samples N_SAMPLES. The sampled data will be saved automatically and used in the following experiments.

   NOTE: Sampled will be used in every run. Delete /data/data1k if you want to use different set.

4. Tune c1: --run tune_c1

   Reproduce Supplementary Figure 2a: Hyperparameter Tuning c1.

5. Explore dataset size: --run dataset_size

   Reproduce Figure 3: Learning Curve and Sample Variance.

6. Build data280: --run data280

   Select 280 data and compute kernel on these data.

7. Tune C_noiseless: --run tune_Cnl

   Reproduce Supplementary Figure 2b: Hyperparameter Tuning C on noiseless circuit.

8. Tune C_hardware: --run tune_Chw

   Reproduce Supplementary Figure 8: Hyperparameter Tuning C on hardware with noise.

9. Noiseless vs. Experimental Results: --run NLvsHW

   Reproduce Supplementary Figure 4c: Noiseless vs. Experimental Results.

pip install -r requirements.txt

Only test on cpu with Schwimmbad library to accelerate processing and computing, not sure whether it will conflict with paddlepaddle-gpu.

• Number of features

  67 VS. 53

  53 features:

  	feature		feature		feature		feature
  1	n_measure	16	sum_flux_err2	31	fou2_0	46	fou1_4
  2	min_flux	17	skew_flux_err2	32	kur_0	47	fou2_4
  3	max_flux	18	mean_interval	33	skew_0	48	kur_4
  4	mean_flux	19	max_interval	34	fou1_1	49	skew_4
  5	med_flux	20	hostgal_specz	35	fou2_1	50	fou1_5
  6	std_flux	21	hostgal_photoz	36	kur_1	51	fou2_5
  7	skew_flux	22	hostgal_photoz_err	37	skew_1	52	kur_5
  8	min_flux_err	23	ra	38	fou1_2	53	skew_5
  9	max_flux_err	24	decl	39	fou2_2
  10	mean_flux_err	25	gal_l	40	kur_2
  11	med_flux_err	26	gal_b	41	skew_2
  12	std_flux_err	27	ddf	42	fou1_3
  13	skew_flux_err	28	distmod	43	fou2_3
  14	sum_flux_err_ratio	29	mwebv	44	kur_3
  15	skew_flux_err_ratio	30	fou1_0	45	skew_3

• Fourier coefficient

  Omit imaginary part?

• Logscale

  logscale_features = ['min_flux', 'max_flux', 'mean_flux', 'med_flux', 'std_flux', 'skew_flux',\
                      'min_flux_err', 'max_flux_err', 'mean_flux_err', 'med_flux_err', 'std_flux_err',\
                      'skew_flux_err', 'sum_flux_err_ratio', 'skew_flux_err_ratio', 'sum_flux_err2',\
                      'skew_flux_err2', 'hostgal_specz', 'hostgal_photoz', 'hostgal_photoz_err']

• Robust scaler

  Transformation in paper: $$x_k'=\pi (\frac{x_k-P_1}{P_{99}-P_1})-\frac{\pi}{2}$$ is different from sklearn.preprocessing.RobustScaler.