add models and scripts

adv/2dplanes/best_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50550fb39b33f5b915793c3bb5ef65221cacf0231eb58383293b778626c18149
+size 292683

adv/MiniBooNE/best_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4651ec0a6994d39c0615c3a914a3e1953f8bf916ddce5eaf34103f5ef8df846
+size 333643

adv/electricity/best_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24a51559d8b04472596dc323f0fa12b9d9bc6157f673fc0b2143032c8095159e
+size 288587

adv/fried/best_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7acd9b0fdbaefc284b3973e8d2e317a9838f038e22b753c0cdef7fb2128cf1e
+size 292683

adv/nomao/best_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:114cb75c22580e8a731eafa11cf67fb721022748500e389763bd67d6de49e08e
+size 373579

adv/pol/best_checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:229d38de571687b49de6ce6567d280cbd40a2d469b657463d8cf7238d4ad7913
+size 331595

fastshap/2dplanes/explainer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d972175f11b885ca3643b14afc2e42b00284a8164ac02157901e73b60d2ec3ee
+size 86108

fastshap/2dplanes/imputer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1a51cdc977bd70242b643051ec22e4e4e9b16ac31d44cba52cd4075a0df88da
+size 82120

fastshap/MiniBooNE/explainer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5b6f3116ce7073f36e35b9515d6bd5fb2feff38013264127e435a3eed542cac
+size 147868

fastshap/MiniBooNE/imputer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65b6789c8d86f1247ec8ecab4edc6db4e3225ab9d2ce472adff6984e746852b2
+size 123272

fastshap/electricity/explainer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:730a7c6d948db8617b336958a26eb67cd78cc791faef92313324c4e6909c97b3
+size 79900

fastshap/electricity/imputer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f3031d87212634d28af9fb43c1d7b71a2c9d8c51a9364d28d03c0b4accf38eb
+size 78024

fastshap/fried/explainer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:372c92f480bcaf28595c17e7608a53b38eebc82310171af0a7be95a8dba28ea6
+size 86108

fastshap/fried/imputer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1965b76e191feff4b2c24a42d0f351a0616de0e77aa45aa18098e7039ab3caf3
+size 82056

fastshap/nomao/explainer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43b7e5290a2973240082853ba9f195949e7be96cef0d784b5c432941c6bda1ae
+size 208092

fastshap/nomao/imputer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a2497dd4411343026cdab089ef238a27c988b36143cdfe1c9274de98e7279e5
+size 162952

fastshap/pol/explainer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e602a98b32f2d4066b60b709444031a4ffd3cdc4bbee4281b4fbd8e810374280
+size 144796

fastshap/pol/imputer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34bcd404134aab210b84849c6bace214b7f1577c3082b3ef7d7d3c046c8e7627
+size 120968

use_adv.py

@@ -0,0 +1,56 @@
+import torch
+import os
+import numpy as np
+
+import lightning as pl
+from torch.utils.data import TensorDataset, DataLoader
+from adv import ValuationModel
+from opendataval.dataloader import DataFetcher
+
+Array = np.ndarray
+
+def load_model(dataset_name: str) -> ValuationModel:
+    results = torch.load(
+        os.path.join("adv", dataset_name, "best_checkpoint.pt")
+    )
+    return results
+
+def onehot_to_location(y: Array) -> Array:
+    res = np.zeros((y.shape[0]), dtype=int)
+    for i in range(y.shape[0]):
+        for j in range(len(y[i])):
+            if y[i, j] > 0:
+                res[i] = j
+                break
+    return res
+
+def load_dataset(dataset_name: str, train_size: int, valid_size: int) -> tuple[Array, Array, Array, Array]:
+    fetcher = DataFetcher(dataset_name=dataset_name).split_dataset_by_count(
+        train_size, valid_size
+    )
+    x_train, y_train, x_val, y_val, _, _ = fetcher.datapoints
+    x_train = np.array(x_train.tolist())
+    y_train = onehot_to_location(np.array(y_train.tolist()))
+    x_val = np.array(x_val.tolist())
+    y_val = onehot_to_location(np.array(y_val.tolist()))
+    return x_train, y_train, x_val, y_val
+
+checkpoint = load_model("electricity")
+x_train, y_train, *_ = load_dataset("electricity", 100, 0)
+tensor_x = torch.Tensor([x_train[24]])
+tensor_y = torch.Tensor([y_train[24]])
+tensor_y = tensor_y.long()
+
+dataset = TensorDataset(tensor_x, tensor_y)
+dataloader = DataLoader(dataset, shuffle=False, drop_last=False, num_workers=0)
+trainer = pl.Trainer(
+    precision="bf16-mixed", accelerator="gpu", devices=[0], enable_progress_bar=False
+)
+estimates = (
+    torch.cat(trainer.predict(checkpoint, dataloader))
+    .squeeze()
+    .cpu()
+    .float()
+)
+
+print(estimates.numpy())

use_fastshap.py

@@ -0,0 +1,39 @@
+import torch
+import torch.nn as nn
+import os
+import numpy as np
+
+from fastshap import FastSHAP
+from opendataval.dataloader import DataFetcher
+
+Array = np.ndarray
+
+def load_model(dataset_name: str) -> FastSHAP:
+    explainer = torch.load(os.path.join("fastshap", dataset_name, "explainer.pt"))
+    imputer = torch.load(os.path.join("fastshap", dataset_name, "imputer.pt"))
+    shap = FastSHAP(explainer, imputer, link=nn.Softmax(dim=-1))
+    return shap
+
+def onehot_to_location(y: Array) -> Array:
+    res = np.zeros((y.shape[0]))
+    for i in range(y.shape[0]):
+        for j in range(len(y[i])):
+            if y[i, j] > 0:
+                res[i] = j
+                break
+    return res
+
+def load_dataset(dataset_name: str, train_size: int, valid_size: int) -> tuple[Array, Array, Array, Array]:
+    fetcher = DataFetcher(dataset_name=dataset_name).split_dataset_by_count(
+        train_size, valid_size
+    )
+    x_train, y_train, x_val, y_val, _, _ = fetcher.datapoints
+    x_train = np.array(x_train.tolist())
+    y_train = onehot_to_location(np.array(y_train.tolist()))
+    x_val = np.array(x_val.tolist())
+    y_val = onehot_to_location(np.array(y_val.tolist()))
+    return x_train, y_train, x_val, y_val
+
+fastshap = load_model("electricity")
+x_train, y_train, *_ = load_dataset("electricity", 100, 0)
+print(np.sum(fastshap.shap_values(np.array([x_train[24]]))[:, int(y_train[24])]))