FCD from fcd_torch (molecularsets#29)

* FCD from fcd_torch

* evaluation script gpu -> device

* pyTorch 1.0.1

* shm-size

* save config and vocab first

* jtnn sampling fix

* device in run.py

* minor fixes

* minor fixes

* share pool

* share pool

* close pool

Dockerfile

@@ -30,9 +30,8 @@ RUN set -ex \
 
 ENV PATH /opt/miniconda/bin:$PATH
 RUN conda install -yq numpy=1.15.0 scipy=1.1.0 matplotlib=3.0.1 pandas=0.23.3 scikit-learn=0.19.1 tqdm \
-    && pip install tensorflow-gpu==1.12 keras==2.2.4 \
     && conda install -yq -c rdkit rdkit=2018.09.1.0 \
-    && conda install -yq -c pytorch pytorch=0.4.1 torchvision=0.2.1 \
+    && conda install -yq -c pytorch pytorch=1.0.1 torchvision=0.2.1 \
     && conda clean -yq -a
 
 WORKDIR /moses

README.md

@@ -165,7 +165,7 @@ nvidia-docker image build --tag molecularsets/moses moses/
 
 3. Create a container:
 ```
-nvidia-docker run -it --name moses --network="host" --shm-size 1G molecularsets/moses
+nvidia-docker run -it --name moses --network="host" --shm-size 10G molecularsets/moses
 ```
 
 4. The dataset and source code are available inside the docker container at /moses:
@@ -250,9 +250,9 @@ python scripts/run.py
 ```
 This will **split** the dataset, **train** the models, **generate** new molecules, and **calculate** the metrics. Evaluation results will be saved in `metrics.csv`.
 
-You can specify the GPU index (-1 for CPU) and/or model by running:
+You can specify the GPU device index as `cuda:n` (or `cpu` for CPU) and/or model by running:
 ```
-python scripts/run.py --gpu 1 --model aae
+python scripts/run.py --device cuda:1 --model aae
 ```
 
 For more details run `python scripts/run.py --help`.

moses/__init__.py

@@ -1 +1 @@
-__version__ = '0.1.0'
+__version__ = '0.1.1'

moses/junction_tree/jtnn/jtnn_dec.py

@@ -218,7 +218,10 @@ def decode(self, mol_vec, prob_decode):
 
                 if prob_decode:
                     b = pred_score.data.squeeze().cpu()  # TODO
-                    sort_wid = torch.multinomial(b, 5)
+                    to_sample = 5
+                    nonzero = (b != 0).long().sum()
+                    to_sample = min(to_sample, nonzero)
+                    sort_wid = torch.multinomial(b, to_sample)
                 else:
                     _, sort_wid = torch.sort(pred_score, dim=1, descending=True)
                     sort_wid = sort_wid.data.squeeze()

moses/junction_tree/jtnn/jtnn_vae.py

@@ -276,6 +276,8 @@ def dfs_assemble(self, tree_mess, mol_vec, all_nodes, cur_mol, global_amap, fa_a
 
         if prob_decode:
             probs = nn.Softmax(dim=-1)(scores.view(1, -1)).squeeze() + 1e-5  # prevent prob = 0
+            if probs.ndimension() == 0:
+                probs = probs[None,]
             cand_idx = torch.multinomial(probs, probs.numel())
         else:
             _, cand_idx = torch.sort(scores, descending=True)

moses/metrics/NP_Score/npscorer.py

@@ -1,7 +1,8 @@
 #
 # calculation of natural product-likeness as described in:
 #
-# Natural Product-likeness Score and Its Application for Prioritization of Compound Libraries
+# Natural Product-likeness Score and Its Application for Prioritization of
+# Compound Libraries
 # Peter Ertl, Silvio Roggo, and Ansgar Schuffenhauer
 # Journal of Chemical Information and Modeling, 48, 68-74 (2008)
 # http://pubs.acs.org/doi/abs/10.1021/ci700286x
@@ -98,7 +99,9 @@ def processMols(fscore, suppl):
 if __name__ == '__main__':
     fscore = readNPModel()  # fills fscore
 
-    suppl = Chem.SmilesMolSupplier(sys.argv[1], smilesColumn=0, nameColumn=1, titleLine=False)
+    suppl = Chem.SmilesMolSupplier(
+        sys.argv[1], smilesColumn=0, nameColumn=1, titleLine=False
+    )
     processMols(fscore, suppl)
 
 #
@@ -117,7 +120,8 @@ def processMols(fscore, suppl):
 #       with the distribution.
 #     * Neither the name of Novartis Institutes for BioMedical Research Inc.
 #       nor the names of its contributors may be used to endorse or promote
-#       products derived from this software without specific prior written permission.
+#       products derived from this software without specific prior written
+#       permission.
 #
 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT

moses/metrics/SA_Score/sascorer.py

@@ -54,8 +54,9 @@ def calculateScore(m):
         readFragmentScores()
 
     # fragment score
-    fp = rdMolDescriptors.GetMorganFingerprint(m,
-                                               2)  # <- 2 is the *radius* of the circular fingerprint
+    fp = rdMolDescriptors.GetMorganFingerprint(
+        m, 2  # <- 2 is the *radius* of the circular fingerprint
+    )
     fps = fp.GetNonzeroElements()
     score1 = 0.
     nf = 0
@@ -87,7 +88,8 @@ def calculateScore(m):
     if nMacrocycles > 0:
         macrocyclePenalty = math.log10(2)
 
-    score2 = 0. - sizePenalty - stereoPenalty - spiroPenalty - bridgePenalty - macrocyclePenalty
+    score2 = (0. - sizePenalty - stereoPenalty -
+              spiroPenalty - bridgePenalty - macrocyclePenalty)
 
     # correction for the fingerprint density
     # not in the original publication, added in version 1.1
@@ -158,7 +160,8 @@ def processMols(mols):
 #       with the distribution.
 #     * Neither the name of Novartis Institutes for BioMedical Research Inc.
 #       nor the names of its contributors may be used to endorse or promote
-#       products derived from this software without specific prior written permission.
+#       products derived from this software without specific prior written
+#       permission.
 #
 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT