release candidate

Author: stefanch

sgdml/__init__.py

@@ -22,7 +22,7 @@
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 # SOFTWARE.
 
-__version__ = '1.0.0.dev0'
+__version__ = '1.0.0'
 
 MAX_PRINT_WIDTH = 100
 LOG_LEVELNAME_WIDTH = 7  # do not modify
@@ -109,7 +109,7 @@ def __init__(self, name):
         hd.setFormatter(formatter)
         hd.setLevel(
             logging.INFO
-        )  # control logging level here
+        ) # control logging level here
 
         self.addHandler(hd)
         return

sgdml/cli.py

@@ -140,7 +140,6 @@ def _print_splash(max_memory, max_processes, use_torch):
             'You can update your installation by running \'pip install sgdml --upgrade\'.'
         )
 
-
     _print_billboard()
 
 
@@ -187,6 +186,7 @@ def _print_billboard():
     bbs = None
     try:
         import json
+
         bbs = json.loads(resp_str)
     except:
         pass
@@ -1174,7 +1174,7 @@ def _online_err(err, size, n, mae_n_sum, rmse_n_sum):
     mae_n_sum += np.sum(err) / size
     mae = mae_n_sum / n
 
-    rmse_n_sum += np.sum(err ** 2) / size
+    rmse_n_sum += np.sum(err**2) / size
     rmse = np.sqrt(rmse_n_sum / n)
 
     return mae, mae_n_sum, rmse, rmse_n_sum
@@ -1768,7 +1768,7 @@ def test(
                     'rmse': e_rmse.item(),
                 }
 
-            model['f_err'] = {'mae':f_mae.item(), 'rmse': f_rmse.item()}
+            model['f_err'] = {'mae': f_mae.item(), 'rmse': f_rmse.item()}
             np.savez_compressed(model_path, **model)
 
             if is_test and model['n_test'] > 0:
@@ -1994,7 +1994,7 @@ def _add_argument_dir_with_file_type(parser, type, or_file=False):
         )
 
     # Available resources
-    total_memory = psutil.virtual_memory().total // 2 ** 30
+    total_memory = psutil.virtual_memory().total // 2**30
     total_cpus = mp.cpu_count()
 
     parser = argparse.ArgumentParser()

sgdml/get.py

@@ -2,7 +2,7 @@
 
 # MIT License
 #
-# Copyright (c) 2018-2021 Stefan Chmiela
+# Copyright (c) 2018-2023 Stefan Chmiela
 #
 # Permission is hereby granted, free of charge, to any person obtaining a copy
 # of this software and associated documentation files (the "Software"), to deal
@@ -131,15 +131,15 @@ def main():
     print()
     print('Available %ss:' % args.command)
 
-    print('{:<2} {:<25}    {:>4}'.format('ID', 'Name', 'Size'))
-    print('-' * 36)
+    print('{:<2} {:<31}    {:>4}'.format('ID', 'Name', 'Size'))
+    print('-' * 42)
 
     items = line[0].split(b';')
     for i, item in enumerate(items):
         name, size = item.split(b',')
-        size = int(size) / 1024 ** 2  # Bytes to MBytes
+        size = int(size) / 1024**2  # Bytes to MBytes
 
-        print('{:>2d} {:<24} {:>5.1f} MB'.format(i, name.decode("utf-8"), size))
+        print('{:>2d} {:<30} {:>5.1f} MB'.format(i, name.decode("utf-8"), size))
     print()
 
     down_list = raw_input(
@@ -155,12 +155,12 @@ def main():
         down_idxs = [int(idx) for idx in re.split(r'\s+', down_list.strip())]
         down_idxs = list(set(down_idxs))
     else:
-        print(ui.warn_str('ABORTED'))
+        print(ui.color_str('ABORTED', fore_color=ui.RED, bold=True))
 
     for idx in down_idxs:
         if idx not in range(len(items)):
             print(
-                ui.warn_str('[WARN]')
+                ui.color_str('[WARN]', fore_color=ui.YELLOW, bold=True)
                 + ' Index '
                 + str(idx)
                 + ' out of range, skipping.'

sgdml/predict.py

@@ -172,7 +172,7 @@ def _predict_wkr(
 
     # avoid divisions (slower)
     sig_inv = 1.0 / sig
-    mat52_base_fact = 5.0 / (3 * sig ** 3)
+    mat52_base_fact = 5.0 / (3 * sig**3)
     diag_scale_fact = 5.0 / sig
     sqrt5 = np.sqrt(5.0)
 
@@ -313,7 +313,7 @@ def __init__(
         if log_level is not None:
             self.log.setLevel(log_level)
 
-        total_memory = psutil.virtual_memory().total // 2 ** 30  # bytes to GB)
+        total_memory = psutil.virtual_memory().total // 2**30  # bytes to GB)
         self.max_memory = (
             min(max_memory, total_memory) if max_memory is not None else total_memory
         )
@@ -378,14 +378,14 @@ def __init__(
                 self.torch_predict = torch.nn.DataParallel(self.torch_predict)
 
             # Send model to device
-            #self.torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
+            # self.torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
             if _torch_cuda_is_available:
                 self.torch_device = 'cuda'
             elif _torch_mps_is_available:
                 self.torch_device = 'mps'
             else:
                 self.torch_device = 'cpu'
-            
+
             while True:
                 try:
                     self.torch_predict.to(self.torch_device)
@@ -405,9 +405,9 @@ def __init__(
                             model.set_n_perm_batches(
                                 model.get_n_perm_batches() + 1
                             )  # uncache
-                            #self.torch_predict.to( # NOTE!
+                            # self.torch_predict.to( # NOTE!
                             #    self.torch_device
-                            #)  # try sending to device again
+                            # )  # try sending to device again
                             pass
                         else:
                             self.log.critical(
@@ -1194,12 +1194,16 @@ def predict(self, R=None, return_E=True):
                     print()
                     os._exit(1)
             else:
-                R_torch = torch.from_numpy(R.reshape(-1, self.n_atoms, 3)).type(torch.float32).to(
-                    self.torch_device
+                R_torch = (
+                    torch.from_numpy(R.reshape(-1, self.n_atoms, 3))
+                    .type(torch.float32)
+                    .to(self.torch_device)
                 )
 
             model = self.torch_predict
-            if R_torch.shape[0] < torch.cuda.device_count() and isinstance(model, torch.nn.DataParallel):
+            if R_torch.shape[0] < torch.cuda.device_count() and isinstance(
+                model, torch.nn.DataParallel
+            ):
                 model = self.torch_predict.module
             E_torch_F_torch = model.forward(R_torch, return_E=return_E)
 

sgdml/solvers/iterative.py

@@ -148,7 +148,7 @@ def _init_kernel_operator(
         n_train = R_desc.shape[0]
 
         # dummy alphas
-        v_F = np.zeros((n-n_train, 1)) if task['use_E_cstr'] else np.zeros((n, 1))
+        v_F = np.zeros((n - n_train, 1)) if task['use_E_cstr'] else np.zeros((n, 1))
         v_E = np.zeros((n_train, 1)) if task['use_E_cstr'] else None
 
         # Note: The standard deviation is set to 1.0, because we are predicting normalized labels here.
@@ -372,12 +372,16 @@ def _lev_scores(
         dim_m = dim_i * min(n_inducing_pts, 10)
 
         # Which columns to use for leverage score approximation?
-        lev_approx_idxs = np.sort(np.random.choice(n_train*dim_i + (n_train if use_E_cstr else 0), dim_m, replace=False)) # random subset of columns
-        #lev_approx_idxs = np.sort(np.random.choice(n_train*dim_i, dim_m, replace=False)) # random subset of columns
-        
-        #lev_approx_idxs = np.s_[
+        lev_approx_idxs = np.sort(
+            np.random.choice(
+                n_train * dim_i + (n_train if use_E_cstr else 0), dim_m, replace=False
+            )
+        )  # random subset of columns
+        # lev_approx_idxs = np.sort(np.random.choice(n_train*dim_i, dim_m, replace=False)) # random subset of columns
+
+        # lev_approx_idxs = np.s_[
         #    :dim_m
-        #]  # first 'dim_m' columns (faster kernel construction)
+        # ]  # first 'dim_m' columns (faster kernel construction)
 
         L_inv_K_mn = self._nystroem_cholesky_factor(
             R_desc,
@@ -460,7 +464,7 @@ def _cho_factor_stable(self, M, pre_reg=False, eps_mag_max=1):
 
         self.log.critical(
             'Failed to factorize despite strong regularization (max: {})!\nYou could try a larger sigma.'.format(
-                10.0 ** eps_mag_max
+                10.0**eps_mag_max
             )
         )
         print()
@@ -492,7 +496,7 @@ def solve(
         num_iters0 = task['solver_iters'] if 'solver_iters' in task else 0
 
         # Number of inducing points to use for Nystrom approximation.
-        max_memory_bytes = self._max_memory * 1024 ** 3
+        max_memory_bytes = self._max_memory * 1024**3
         max_n_inducing_pts = Iterative.max_n_inducing_pts(
             n_train, n_atoms, max_memory_bytes
         )
@@ -849,13 +853,12 @@ def max_n_inducing_pts(n_train, n_atoms, max_memory_bytes):
         ny_factor = SQUARE_FACT * to_dof
 
         n_inducing_pts = (
-            np.sqrt(sq_factor ** 2 + 4.0 * ny_factor * max_memory_bytes) - sq_factor
+            np.sqrt(sq_factor**2 + 4.0 * ny_factor * max_memory_bytes) - sq_factor
         ) / (2 * ny_factor)
         n_inducing_pts = int(n_inducing_pts)
 
         return min(n_inducing_pts, n_train)
 
-
     @staticmethod
     def est_memory_requirement(n_train, n_inducing_pts, n_atoms):