[12]:

import torch
import matplotlib.pyplot as plt
import irsa
import glob
import numpy as np

torch.cuda.is_available()

[12]:

True

Load and Prepare Data

[18]:

# Paths to experimental data
exp_paths = sorted(glob.glob('data/DX/*.DX'))

# Paths to predicted data
pred_paths = sorted(glob.glob('/qfs/projects/cryo_ir/nist/dft_under10conf/**/*.pkl'))

[3]:

# Load single experimental data instance
x_exp, y_exp = irsa.io.load_experimental(exp_paths[0])

[4]:

# Load single predicted data instance
x_pred_raw, y_pred_raw = irsa.io.load_predicted(pred_paths[0])

[5]:

# Process predicted
x_pred, y_pred = irsa.io.preprocess_predicted(x_pred_raw, y_pred_raw, x_exp, sigma=2)

[6]:

fig, ax = plt.subplots(2, dpi=150, facecolor='w', sharex=True, sharey=True)

# Plot experimental
ax[0].plot(x_exp, y_exp / y_exp.max(), c='C0', label='Exp', alpha=0.5)
ax[1].plot(x_exp, y_exp / y_exp.max(), c='C0', label='Exp', alpha=0.5)


# Plot raw prediction
_, stemlines, _ = ax[0].stem(x_pred_raw, y_pred_raw / y_pred_raw.max(),
                             markerfmt=" ",
                             basefmt=" ",
                             linefmt="C3-",
                             use_line_collection=True,
                             label='Pred')
plt.setp(stemlines, 'alpha', 0.5)

# Plot processed prediction
ax[1].plot(x_pred, y_pred / y_pred.max(), c='C3', label='Pred', alpha=0.5)

# Limits
ax[0].set_ylim(0, None)

# Axis labels
ax[0].set_ylabel('Intensity', fontweight='bold')
ax[1].set_xlabel('Wavenumber', fontweight='bold')
ax[1].set_ylabel('Intensity', fontweight='bold')

# Legend
ax[0].legend()

plt.tight_layout()
plt.show()
../_images/user_guide_Scoring_6_0.png

Load Model

[7]:

# Load model
model = irsa.io.load_model('model/ir_exp_pred_pnn_0.1757.pt', embedding_dim=2048)
model

[7]:

PairedNeuralNet(
  (domain_embed1): DomainEncoder(
    (conv1): Conv1d(1, 32, kernel_size=(21,), stride=(1,), padding=same)
    (conv2): Conv1d(32, 32, kernel_size=(21,), stride=(1,), padding=same)
    (conv3): Conv1d(32, 64, kernel_size=(21,), stride=(1,), padding=same)
    (conv4): Conv1d(64, 64, kernel_size=(21,), stride=(1,), padding=same)
    (bn1): BatchNorm1d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (bn2): BatchNorm1d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (bn3): BatchNorm1d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (bn4): BatchNorm1d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (fc_out): Linear(in_features=3264, out_features=2048, bias=True)
  )
  (domain_embed2): DomainEncoder(
    (conv1): Conv1d(1, 32, kernel_size=(21,), stride=(1,), padding=same)
    (conv2): Conv1d(32, 32, kernel_size=(21,), stride=(1,), padding=same)
    (conv3): Conv1d(32, 64, kernel_size=(21,), stride=(1,), padding=same)
    (conv4): Conv1d(64, 64, kernel_size=(21,), stride=(1,), padding=same)
    (bn1): BatchNorm1d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (bn2): BatchNorm1d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (bn3): BatchNorm1d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (bn4): BatchNorm1d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (fc_out): Linear(in_features=3264, out_features=2048, bias=True)
  )
  (bn1): BatchNorm1d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (fc_out): Linear(in_features=2048, out_features=1, bias=True)
)

Predict

[8]:

irsa.predict.predict(model, y_exp, y_pred)

[8]:

array([0.99995506], dtype=float32)

More than one comparison

[19]:

# Experimental
x_exp, y_exp = zip(*[irsa.io.load_experimental(x) for x in exp_paths[:100]])
x_exp = np.array(x_exp)
y_exp = np.array(y_exp)

[20]:

# Predicted
x_pred, y_pred = zip(*[irsa.io.preprocess_predicted(*irsa.io.load_predicted(x), x_exp[0]) for x in pred_paths[:100]])
x_pred = np.array(x_pred)
y_pred = np.array(y_pred)

[21]:

irsa.predict.predict(model, y_exp, y_pred)

[21]:

array([9.98715878e-01, 9.99961257e-01, 8.05124164e-01, 2.83133658e-03,
       5.09623169e-06, 1.73197100e-06, 9.99685526e-01, 1.96168621e-04,
       4.47442057e-04, 8.02679278e-06, 4.69562224e-07, 5.04134259e-05,
       6.25733321e-07, 5.60668650e-06, 3.56422141e-02, 4.65421326e-04,
       2.02869391e-03, 3.01517343e-06, 3.65227720e-07, 6.11377686e-08,
       1.19881440e-06, 3.43599617e-01, 1.61735285e-02, 5.78778327e-07,
       3.71149153e-01, 1.57151022e-03, 8.28673365e-05, 1.43781293e-03,
       2.97027273e-05, 3.69805425e-01, 1.41901939e-04, 6.47961974e-01,
       4.82831821e-02, 1.41528547e-01, 7.23487290e-07, 1.24725675e-05,
       1.03870351e-02, 8.09177533e-02, 3.51061864e-08, 2.60172305e-06,
       1.00584212e-03, 1.46003149e-05, 2.41527755e-06, 1.70100422e-03,
       8.38411099e-04, 6.15208119e-05, 8.95332396e-02, 5.66744924e-01,
       6.33308897e-04, 2.71203872e-02, 5.56705929e-02, 8.25917125e-01,
       9.62753892e-02, 6.67654335e-01, 4.72425343e-03, 9.42284714e-06,
       1.18774265e-01, 8.26731321e-06, 6.16460025e-01, 1.30764619e-01,
       2.68799081e-08, 4.97228408e-04, 3.85968560e-05, 2.25130225e-05,
       4.80837487e-02, 1.33044575e-09, 7.21884930e-07, 1.47229744e-04,
       2.77990307e-06, 3.17709464e-06, 1.55754564e-02, 2.15555236e-01,
       2.13376489e-02, 1.85871511e-04, 4.14195483e-06, 5.08218771e-04,
       7.59210289e-02, 3.69078742e-04, 2.21808295e-05, 4.51831170e-07,
       7.84295380e-01, 3.38352323e-01, 5.96173584e-01, 3.74672413e-02,
       4.88443902e-06, 1.93366868e-05, 7.50493700e-06, 7.49912761e-06,
       1.80096515e-02, 6.17892966e-02, 9.80744779e-01, 6.14370365e-05,
       1.68816905e-06, 7.10638778e-06, 3.36672315e-06, 1.24060762e-05,
       4.70419800e-05, 4.89540753e-06, 3.73172313e-02, 1.24584453e-03],
      dtype=float32)

From already processed data

[22]:

exp = np.load('data/exp.npy')
pred = np.load('data/pred.npy')
exp_labels = np.load('data/exp_labels.npy')
pred_labels = np.load('data/pred_labels.npy')

[23]:

irsa.predict.predict(model, exp[:100], pred[:100])

[23]:

array([9.9806887e-01, 9.8847932e-01, 9.7819775e-01, 1.1839760e-04,
       1.8524046e-05, 9.9987423e-01, 2.0489060e-05, 1.6157774e-08,
       2.6285807e-03, 2.5301302e-04, 1.4667533e-02, 6.6556209e-01,
       1.1246317e-09, 1.1485731e-07, 5.8699306e-04, 2.0577816e-02,
       1.9643754e-03, 1.4899313e-05, 1.8422359e-05, 7.5439243e-03,
       1.7297349e-05, 5.1432671e-03, 2.3997613e-06, 1.4830708e-04,
       1.5003191e-07, 1.7636169e-02, 9.3209529e-03, 1.5387259e-02,
       3.6595590e-02, 6.2160776e-05, 1.0429835e-06, 7.5525844e-05,
       3.1944439e-01, 2.7835724e-05, 1.5231686e-06, 9.3302197e-07,
       6.6279041e-05, 1.8409251e-01, 7.8748119e-11, 1.8406306e-06,
       1.3092839e-03, 2.4196327e-06, 1.9472660e-04, 1.8026882e-04,
       5.7782349e-04, 1.5409833e-01, 2.3946088e-02, 4.1309467e-01,
       7.2982555e-05, 5.8766623e-04, 1.5705214e-06, 7.1283258e-02,
       3.5826121e-02, 4.3652490e-01, 1.0370320e-04, 8.2009058e-07,
       5.0271320e-04, 1.0014056e-03, 3.4454702e-03, 7.7238502e-07,
       1.9698977e-04, 1.6836447e-06, 4.5183808e-07, 2.3351687e-01,
       3.1644895e-03, 1.2672196e-02, 5.5354819e-02, 1.6746003e-07,
       6.6230565e-10, 1.3943630e-07, 2.0257197e-03, 3.5122088e-05,
       9.2617908e-10, 9.9689424e-01, 1.2731661e-07, 2.3909392e-05,
       3.7010829e-04, 3.4838724e-01, 9.0574908e-01, 2.0915195e-06,
       2.3226252e-05, 1.1664160e-02, 8.4716501e-03, 3.3030989e-05,
       1.3482069e-03, 3.2849935e-01, 2.0340875e-04, 2.0326566e-04,
       6.2182366e-06, 2.6787993e-06, 1.5131337e-06, 8.7273382e-02,
       7.1432921e-10, 1.3340642e-03, 8.8962643e-10, 9.1723091e-04,
       2.0955320e-05, 7.0164050e-03, 3.3119373e-02, 2.1554666e-07],
      dtype=float32)

[ ]: