modify ff.py file and add pre-trained folder.

alignn/ff/revised/config.json → alignn/pre-trained/ocp2020_all/config.json

@@ -7,58 +7,52 @@
     "id_tag": "jid",
     "random_seed": 123,
     "classification_threshold": null,
-    "n_val": null,
-    "n_test": null,
-    "n_train": null,
-    "train_ratio": 0.9,
-    "val_ratio": 0.05,
-    "test_ratio": 0.05,
+    "n_val": 24943,
+    "n_test": 24943,
+    "n_train": 460328,
+    "train_ratio": 0.8,
+    "val_ratio": 0.1,
+    "test_ratio": 0.1,
     "target_multiplication_factor": null,
-    "epochs": 100,
-    "batch_size": 16,
+    "epochs": 50,
+    "batch_size": 32,
     "weight_decay": 1e-05,
     "learning_rate": 0.001,
     "filename": "sample",
     "warmup_steps": 2000,
-    "criterion": "l1",
+    "criterion": "mse",
     "optimizer": "adamw",
     "scheduler": "onecycle",
     "pin_memory": false,
     "save_dataloader": false,
     "write_checkpoint": true,
     "write_predictions": true,
-    "store_outputs": false,
+    "store_outputs": true,
     "progress": true,
     "log_tensorboard": false,
     "standard_scalar_and_pca": false,
-    "use_canonize": false,
+    "use_canonize": true,
     "num_workers": 0,
     "cutoff": 8.0,
     "max_neighbors": 12,
-    "keep_data_order": false,
+    "keep_data_order": true,
     "normalize_graph_level_loss": false,
     "distributed": false,
     "n_early_stopping": null,
-    "output_dir": "out_continue",
+    "output_dir": "tempall",
     "model": {
-        "name": "alignn_atomwise",
-        "alignn_layers": 4,
+        "name": "alignn",
+        "alignn_layers": 2,
         "gcn_layers": 4,
         "atom_input_features": 92,
         "edge_input_features": 80,
         "triplet_input_features": 40,
         "embedding_features": 64,
         "hidden_features": 256,
         "output_features": 1,
-        "grad_multiplier": -1,
-        "calculate_gradient": true,
-        "atomwise_output_features": 0,
-        "graphwise_weight": 1.0,
-        "gradwise_weight": 10.0,
-        "stresswise_weight": 0.0,
-        "atomwise_weight": 0.0,
         "link": "identity",
         "zero_inflated": false,
-        "classification": false
+        "classification": false,
+        "num_classes": 2
     }
-}
+}

alignn/pre-trained/ocp2020_all/history_train.json

@@ -0,0 +1 @@
+{"loss": [1.242798488008342, 1.093458558720021, 1.0134414646332985, 0.951067599713243, 0.8745518742396594, 0.898653789754953, 0.8362365446863051, 0.8224021468760861, 0.7795552550399721, 0.7730100799443865, 0.7515591056221759, 0.7388777779588113, 0.716031103362878, 0.7136948209940911, 0.7261467566041014, 0.6732144214893987, 0.6532027991397289, 0.63671875, 0.6265350734271811, 0.6092405826381647, 0.5846296462243656, 0.5580424012100278, 0.5445762933937262, 0.5313807514337852, 0.4952561872718978, 0.47810256316258254, 0.4577896897810219, 0.4254376710766423, 0.40602780402763294, 0.38824577875173794, 0.36579682340545705, 0.34168839068474105, 0.31409366038842546, 0.2938517973909454, 0.26776810832681613, 0.25274615349104973, 0.22926905807373132, 0.20771340113942474, 0.1936585215176399, 0.17776635003910324, 0.16531178378627912, 0.15266229538690476, 0.14320146854362184, 0.13564101806895204], "mae": [0.7805721427485227, 0.7108072916666667, 0.6873055021506778, 0.6571108821037539, 0.6313307183046576, 0.6484085799009385, 0.6195709642205423, 0.6060081845238096, 0.5954230888295099, 0.5953284535757734, 0.5896742347714633, 0.5823936337765033, 0.573410209202294, 0.5723786577815433, 0.5791733875347584, 0.555680525232447, 0.5490652562347932, 0.5480471804940041, 0.5465672657064651, 0.5360462219217066, 0.5294444286040146, 0.5187364564324818, 0.5157378970064304, 0.5108121456269552, 0.49474927088764337, 0.4873516996328641, 0.47867342962504345, 0.46357377748088285, 0.4568008396332986, 0.44969644079118876, 0.43762644057394856, 0.42328173200165103, 0.4096803242852798, 0.3972902681830031, 0.38103306531326037, 0.37213606159845325, 0.3561765117201077, 0.34038708805830725, 0.33032418344847064, 0.31716869514685436, 0.30663672146550225, 0.2952301184502085, 0.286680054092805, 0.2790509353547532]}

alignn/pre-trained/ocp2020_all/history_val.json

@@ -0,0 +1 @@
+{"loss": [1.2597084290255938, 1.1104618018643695, 1.0429591128700657, 0.9956854647635189, 0.9269066003841062, 0.9477316546654364, 0.9062190514933007, 0.9093923311637516, 0.8623399452924422, 0.8719415823822609, 0.8572578087391688, 0.8553321848172738, 0.8403540477826139, 0.8580992163681402, 0.8637968756268052, 0.8405780522905969, 0.8313192776446165, 0.8366862188101734, 0.8379199281430519, 0.8410041231245988, 0.8346832627968549, 0.8368975305128771, 0.8294504579438784, 0.8397048342977776, 0.8445738570834002, 0.8382498627296614, 0.8487222889421534, 0.8316056492698973, 0.8399800801307766, 0.8383183411976493, 0.8575983206635109, 0.8469434941381178, 0.8415695797647224, 0.8477357759096197, 0.8376467194269496, 0.8512415757381258, 0.8520017337431803, 0.8489415924111441, 0.8490380420611361, 0.838059078897224, 0.8546613465530728, 0.8441100212361602, 0.8518101664042442, 0.848463183343028], "mae": [0.7875224396261232, 0.7159629106827664, 0.6968473892309852, 0.6713935352565389, 0.6488075501193437, 0.6651109100459323, 0.6430156350288831, 0.6328940630265164, 0.6233489167552552, 0.6253252335476172, 0.6230350832271944, 0.6184036587873074, 0.6099239246530006, 0.6156974273407413, 0.6214085629111842, 0.6077242301578546, 0.6040773991595796, 0.6076087804752688, 0.608113397835767, 0.6042435025373074, 0.6057539855408577, 0.6031434202377848, 0.6023897653366697, 0.6057553958525553, 0.6024301159213535, 0.5975024476743019, 0.6010299506328225, 0.596541751180901, 0.5968718816441351, 0.5997773979686497, 0.6019575439766528, 0.5958905789298179, 0.5967279906756459, 0.5944703166995948, 0.5925527628946365, 0.5984745833625842, 0.5966328338113768, 0.5940972500802311, 0.5976680417728056, 0.5939875983457357, 0.599169945380195, 0.5955705165250923, 0.5989375965280006, 0.5971947254969312]}

alignn/pre-trained/ocp2020_all/mad

@@ -0,0 +1,3 @@
+MAX val:9.999100859999942
+MIN val:-9.992999119999922
+MAD val:1.750781823274932

alignn/pre-trained/ocp2020_all/run_100k_pre.py

@@ -0,0 +1,203 @@
+# /wrk/knc6/oc/oc2/ocp/data/val_cgcnn_test/pred2.py
+# conda activate ocp-models
+from jarvis.core.atoms import Atoms
+from jarvis.core.specie import atomic_numbers_to_symbols
+
+# from ocpmodels.datasets import SinglePointLmdbDataset
+import os, torch
+from ase.io import read
+
+# from ocpmodels.preprocessing import AtomsToGraphs
+# from ocpmodels.models import CGCNN
+from torch.nn.parallel import DistributedDataParallel as DDP
+import torch.distributed as dist
+from jarvis.core.atoms import Atoms
+
+# from ocpmodels.datasets import data_list_collater
+from tqdm import tqdm
+from jarvis.db.figshare import get_request_data
+import json, zipfile
+import numpy as np
+import pandas as pd
+from jarvis.db.jsonutils import loadjson, dumpjson
+import os
+from jarvis.core.atoms import Atoms
+from jarvis.core.atoms import Atoms
+from jarvis.core.specie import atomic_numbers_to_symbols
+import os, torch
+from ase.io import read
+from torch.nn.parallel import DistributedDataParallel as DDP
+import torch.distributed as dist
+from jarvis.core.atoms import Atoms
+from alignn.models.alignn import ALIGNN, ALIGNNConfig
+import os
+import zipfile
+from tqdm import tqdm
+from alignn.models.alignn import ALIGNN, ALIGNNConfig
+from alignn.data import get_torch_dataset
+from torch.utils.data import DataLoader
+
+dat = get_request_data(js_tag="ocp_all.json",url="https://figshare.com/ndownloader/files/40974599")
+df = pd.DataFrame(dat)
+path = "../../../../../benchmarks/AI/SinglePropertyPrediction/ocp_all_relaxed_energy.json.zip"
+js_tag = "ocp_all_relaxed_energy.json"
+id_data = json.loads(zipfile.ZipFile(path).read(js_tag))
+train_ids = np.array(list(id_data["train"].keys()))
+val_ids = np.array(list(id_data["val"].keys()))
+test_ids = np.array(list(id_data["test"].keys()))
+train_df = df[df["id"].isin(train_ids)]
+val_df = df[df["id"].isin(val_ids)]  # [:take_val]
+test_df = df[df["id"].isin(test_ids)]
+
+print(test_df)
+# https://github.com/Open-Catalyst-Project/ocp/blob/main/configs/is2re/10k/base.yml
+
+device = "cpu"
+if torch.cuda.is_available():
+    device = torch.device("cuda")
+
+model_path = "/wrk/knc6/Software/alignn_calc/jarvis_leaderboard/jarvis_leaderboard/contributions/alignn_model/OCP/all/tempall/checkpoint_45.pt"
+config = "/wrk/knc6/Software/alignn_calc/jarvis_leaderboard/jarvis_leaderboard/contributions/alignn_model/OCP/all/tempall/config.json"
+config_params = loadjson(config)
+
+model = ALIGNN(ALIGNNConfig(name="alignn",alignn_layers=2))
+# model = ALIGNN(ALIGNNConfig(name="alignn",output_features=1,config_params))
+model.load_state_dict(torch.load(model_path, map_location=device)["model"])
+model.to(device)
+model.eval()
+def get_multiple_predictions(
+    model="",
+    atoms_array=[],
+    ids_array=[],
+    cutoff=8,
+    neighbor_strategy="k-nearest",
+    max_neighbors=12,
+    use_canonize=True,
+    target="prop",
+    atom_features="cgcnn",
+    line_graph=True,
+    workers=0,
+    filename="pred_data.json",
+    include_atoms=True,
+    pin_memory=False,
+    output_features=1,
+    batch_size=1,
+    model_name="jv_formation_energy_peratom_alignn",
+    print_freq=100,
+):
+    """Use pretrained model on a number of structures."""
+    # import glob
+    # atoms_array=[]
+    # for i in glob.glob("alignn/examples/sample_data/*.vasp"):
+    #      atoms=Atoms.from_poscar(i)
+    #      atoms_array.append(atoms)
+    # get_multiple_predictions(atoms_array=atoms_array)
+
+    mem = []
+    for i, ii in enumerate(atoms_array):
+        info = {}
+        info["atoms"] = ii.to_dict()
+        info["prop"] = -9999  # place-holder only
+        info["jid"] = str(ids_array[i])
+        mem.append(info)
+
+    # Note cut-off is usually 8 for solids and 5 for molecules
+    def atoms_to_graph(atoms):
+        """Convert structure dict to DGLGraph."""
+        structure = Atoms.from_dict(atoms)
+        return Graph.atom_dgl_multigraph(
+            structure,
+            cutoff=cutoff,
+            atom_features="atomic_number",
+            max_neighbors=max_neighbors,
+            compute_line_graph=True,
+            use_canonize=use_canonize,
+        )
+
+    test_data = get_torch_dataset(
+        dataset=mem,
+        target="prop",
+        neighbor_strategy=neighbor_strategy,
+        atom_features=atom_features,
+        use_canonize=use_canonize,
+        line_graph=line_graph,
+    )
+
+    collate_fn = test_data.collate_line_graph
+    test_loader = DataLoader(
+        test_data,
+        batch_size=batch_size,
+        shuffle=False,
+        collate_fn=collate_fn,
+        drop_last=False,
+        num_workers=workers,
+        pin_memory=pin_memory,
+    )
+
+    results = []
+    with torch.no_grad():
+        ids = test_loader.dataset.ids
+        for dat, id in zip(test_loader, ids):
+            g, lg, target = dat
+            out_data = model([g.to(device), lg.to(device)])
+            out_data = out_data.cpu().numpy().tolist()
+            target = target.cpu().numpy().flatten().tolist()
+            info = {}
+            info["id"] = id
+            info["pred"] = out_data
+            results.append(info)
+            print_freq = int(print_freq)
+            if len(results) % print_freq == 0:
+                print(len(results))
+    df1 = pd.DataFrame(mem)
+    df2 = pd.DataFrame(results)
+    df2["jid"] = df2["id"]
+    df3 = pd.merge(df1, df2, on="jid")
+    save = []
+    for i, ii in df3.iterrows():
+        info = {}
+        info["id"] = ii["id"]
+        info["atoms"] = ii["atoms"]
+        info["pred"] = ii["pred"]
+        save.append(info)
+
+    dumpjson(data=save, filename=filename)
+
+
+# f=open('AI-SinglePropertyPrediction-relaxed_energy-ocp100k-test-mae.csv','w')
+# f.write('id,target,prediction\n')
+# f.write('id,target,scaled_target,prediction\n')
+# print('id,actual,scaled,pred')
+atoms_array = []
+ids_array = []
+for ii, i in tqdm(test_df.iterrows()):
+    fname = i["id"]
+    ids_array.append(fname)
+    atoms = (Atoms.from_dict(i["atoms"]))
+    atoms_array.append(atoms)
+
+    # actual=i['relaxed_energy']
+    # relaxed_energy = (actual-target_mean)/target_std
+    # scaled=relaxed_energy
+    # data = a2g.convert(atoms).to(device)
+    # batch = data_list_collater([data], otf_graph=False)
+    # out = model(batch)
+    # pred=(out[0].detach().cpu().numpy().flatten().tolist()[0])*target_std+target_mean
+    # line=str(fname)+','+str(actual)+','+str(pred) #+'\n'
+    # line=str(i.sid)+','+str(actual)+','+str(scaled)+','+str(pred) #+'\n'
+    # f.write(line+'\n')
+# f.close()
+#atoms_array=atoms_array[0:10]
+#ids_array=ids_array[0:10]
+get_multiple_predictions(
+    model=model, atoms_array=atoms_array, ids_array=ids_array
+)
+from jarvis.db.jsonutils import loadjson
+d=loadjson('pred_data.json')
+f=open('AI-SinglePropertyPrediction-relaxed_energy-ocp_all-test-mae.csv','w')
+f.write('id,prediction\n')
+for i in d:
+ line=i['id']+','+str(i['pred'])+'\n'
+ f.write(line)
+f.close()
+

alignn/pre-trained/ocp2020_all/test_data_data_range

@@ -0,0 +1,2 @@
+Max=9.999100859999942
+Min=-9.975282650000054

alignn/pre-trained/ocp2020_all/train_data_data_range

@@ -0,0 +1,2 @@
+Max=9.963571024999993
+Min=-9.992999119999922

alignn/pre-trained/ocp2020_all/val_data_data_range

@@ -0,0 +1,2 @@
+Max=9.999100859999942
+Min=-9.975282650000054