Loading deepchem/trans/transformers.py +26 −21 Original line number Diff line number Diff line Loading @@ -570,18 +570,19 @@ class CoulombFitTransformer(): class IRVTransformer(): """Performs randomization and binarization operations on batches of Coulomb Matrix features during fit.""" def __init__(self, K, dataset): def __init__(self, K, n_tasks, dataset, transform_X=False, transform_y=False, features=None): """Initializes CoulombFitTransformer. Parameters: ---------- dataset: dc.data.Dataset object dataset: dc.data.Dataset object, y and w should be for one task K: number of nearest neighbours that count K: number of nearest neighbours that count[ """ self.X = dataset.X self.n_samples = X.shape[0] self.n_tasks = n_tasks self.K = K self.y = dataset.y self.w = dataset.w Loading @@ -593,41 +594,45 @@ class CoulombFitTransformer(): "Similarity measurement requires same size") return np.sum(np.min([x,y], axis=0))/np.sum(np.max([x,y], axis=0)) def realize(self, X_target): """Randomize features. """ def _realize_(x_target): def realize(self, X_target, y, w): features = [] for x_target in X_target: similar_xs = [] similar_inds = [] similar_ys = [] threshold = 0 for count, x in enumerate(self.X): if w[count] == 0: continue similar = IRVTransformer.similarity(x, x_target) if similar >= 1: continue if len(similar_xs) < self.K-1: similar_xs.append(similar) similar_inds.append(count) if len(similar_xs) == self.K-1: if self.w[count] similar_ys.append(y[count]) elif len(similar_xs) == self.K-1: similar_xs.append(similar) similar_inds.append(count) similar_ys.append(y[count]) threshold = min(similar_xs) elif similar > threshold: ind = similar_xs.index(threshold) similar_xs.pop(ind) similar_inds.pop(ind) similar_ys.pop(ind) similar_xs.append(similar) similar_inds.append(count) similar_ys.append(y[count]) threshold = min(similar_xs) return similar_xs.extend(similar_inds) return np.array([_realize_(z) for z in X_target]) similar_xs.extend(similar_ys) features.append(similar_xs) return features def X_transform(self, X_target): X_target = self.realize(X_target) return X_target X_target2 = [] for i in range(self.n_tasks): X_target2.append(self.realize(X_target, self.y[:,i], self.w[:,i])) return np.concatenate([z for z in np.array(X_target2)], axis=1) def transform(self, dataset): raise NotImplementedError( "Cannot transform datasets with FitTransformer") X_trans = self.X_transform(dataset.X) return NumpyDataset(X_trans, dataset.y, dataset.w, ids=None) def untransform(self, z): raise NotImplementedError( Loading examples/benchmark.py +1 −1 Original line number Diff line number Diff line Loading @@ -607,7 +607,7 @@ if __name__ == '__main__': for dataset in datasets: if dataset in ['tox21', 'sider', 'muv', 'toxcast', 'pcba', 'clintox']: for model in models: if model in ['tf', 'tf_robust', 'logreg', 'graphconv']: if model in ['tf', 'tf_robust', 'logreg', 'graphconv', 'rf']: benchmark_loading_datasets( hps, dataset=dataset, model=model, split=split, out_path='.') else: Loading Loading
deepchem/trans/transformers.py +26 −21 Original line number Diff line number Diff line Loading @@ -570,18 +570,19 @@ class CoulombFitTransformer(): class IRVTransformer(): """Performs randomization and binarization operations on batches of Coulomb Matrix features during fit.""" def __init__(self, K, dataset): def __init__(self, K, n_tasks, dataset, transform_X=False, transform_y=False, features=None): """Initializes CoulombFitTransformer. Parameters: ---------- dataset: dc.data.Dataset object dataset: dc.data.Dataset object, y and w should be for one task K: number of nearest neighbours that count K: number of nearest neighbours that count[ """ self.X = dataset.X self.n_samples = X.shape[0] self.n_tasks = n_tasks self.K = K self.y = dataset.y self.w = dataset.w Loading @@ -593,41 +594,45 @@ class CoulombFitTransformer(): "Similarity measurement requires same size") return np.sum(np.min([x,y], axis=0))/np.sum(np.max([x,y], axis=0)) def realize(self, X_target): """Randomize features. """ def _realize_(x_target): def realize(self, X_target, y, w): features = [] for x_target in X_target: similar_xs = [] similar_inds = [] similar_ys = [] threshold = 0 for count, x in enumerate(self.X): if w[count] == 0: continue similar = IRVTransformer.similarity(x, x_target) if similar >= 1: continue if len(similar_xs) < self.K-1: similar_xs.append(similar) similar_inds.append(count) if len(similar_xs) == self.K-1: if self.w[count] similar_ys.append(y[count]) elif len(similar_xs) == self.K-1: similar_xs.append(similar) similar_inds.append(count) similar_ys.append(y[count]) threshold = min(similar_xs) elif similar > threshold: ind = similar_xs.index(threshold) similar_xs.pop(ind) similar_inds.pop(ind) similar_ys.pop(ind) similar_xs.append(similar) similar_inds.append(count) similar_ys.append(y[count]) threshold = min(similar_xs) return similar_xs.extend(similar_inds) return np.array([_realize_(z) for z in X_target]) similar_xs.extend(similar_ys) features.append(similar_xs) return features def X_transform(self, X_target): X_target = self.realize(X_target) return X_target X_target2 = [] for i in range(self.n_tasks): X_target2.append(self.realize(X_target, self.y[:,i], self.w[:,i])) return np.concatenate([z for z in np.array(X_target2)], axis=1) def transform(self, dataset): raise NotImplementedError( "Cannot transform datasets with FitTransformer") X_trans = self.X_transform(dataset.X) return NumpyDataset(X_trans, dataset.y, dataset.w, ids=None) def untransform(self, z): raise NotImplementedError( Loading
examples/benchmark.py +1 −1 Original line number Diff line number Diff line Loading @@ -607,7 +607,7 @@ if __name__ == '__main__': for dataset in datasets: if dataset in ['tox21', 'sider', 'muv', 'toxcast', 'pcba', 'clintox']: for model in models: if model in ['tf', 'tf_robust', 'logreg', 'graphconv']: if model in ['tf', 'tf_robust', 'logreg', 'graphconv', 'rf']: benchmark_loading_datasets( hps, dataset=dataset, model=model, split=split, out_path='.') else: Loading
