fixes

.travis.yml

@@ -9,8 +9,10 @@ before_install:
   # version is the same.
   - if [[ "$TRAVIS_PYTHON_VERSION" == "2.7" ]]; then
       wget https://repo.continuum.io/miniconda/Miniconda-latest-Linux-x86_64.sh -O miniconda.sh;
+      export TF_PACKAGE=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.10.1-cp27-none-linux_x86_64.whl
     else
       wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh;
+      export TF_PACKAGE=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.10.1-cp35-cp35m-linux_x86_64.whl
     fi
   - bash miniconda.sh -b -p $HOME/miniconda
   - export PATH="$HOME/miniconda/bin:$PATH"
@@ -29,9 +31,11 @@ addons:
 install:
   - >
       conda create -q -n test-environment python=$TRAVIS_PYTHON_VERSION
-      numpy scipy nose pandas matplotlib mkl-service 
+      numpy scipy nose pandas matplotlib mkl-service theano
   - source activate test-environment
   - pip install tensorflow pypandoc pylint 'theano>=1.0.4'
+  - pip install pypandoc pylint
+  - pip install --ignore-installed --upgrade $TF_PACKAGE
   - pip install -r requirements.txt
   - pip install .
   - pip freeze

mhcflurry/allele_encoding_transforms.py

@@ -25,13 +25,13 @@ class AlleleEncodingTransform(object):
 
         Parameters
         ----------
-        fit : string to DataFrame
+        fit : string to array
         """
 
 
 class PCATransform(AlleleEncodingTransform):
     name = 'pca'
-    serialization_keys = ['data']
+    serialization_keys = ['mean', 'components']
 
     def __init__(self):
         self.model = None
@@ -51,19 +51,15 @@ class PCATransform(AlleleEncodingTransform):
         print("Fit complete in %0.3f sec." % (time.time() - start))
 
     def get_fit(self):
-        df = pandas.DataFrame(self.model.components_)
-        df.columns = ["pca_%s" % c for c in df.columns]
-        df["mean"] = self.model.mean_
         return {
-            'data': df
+            'mean': self.model.mean_,
+            'components': self.model.components_,
         }
 
     def restore_fit(self, fit):
-        assert list(fit) == ['data']
-        data = fit["data"]
         self.model = sklearn.decomposition.PCA()
-        self.model.mean_ = data["mean"].values
-        self.model.components_ = data.drop(columns="mean").values
+        self.model.mean_ = fit["mean"]
+        self.model.components_ = fit["components"]
 
     def transform(self, allele_representations):
         if not self.is_fit():

mhcflurry/class1_affinity_predictor.py

@@ -368,12 +368,9 @@ class Class1AffinityPredictor(object):
             if transform.is_fit():
                 fit_data = transform.get_fit()
                 assert set(fit_data) == set(transform.serialization_keys)
-                for (serialization_key, fit_df) in fit_data.items():
-                    csv_path = join(
-                        models_dir,
-                        "%s.%s.csv" % (transform.name, serialization_key))
-                    fit_df.to_csv(csv_path)
-                    logging.info("Wrote: %s" % csv_path)
+                target_path = join(models_dir, "%s.npz" % transform.name)
+                numpy.savez(target_path, **fit_data)
+                logging.info("Wrote: %s" % target_path)
 
         if self.allele_to_percent_rank_transform:
             percent_ranks_df = None
@@ -447,15 +444,11 @@ class Class1AffinityPredictor(object):
         for transform_name in ALLELE_ENCODING_TRANSFORMS:
             klass = ALLELE_ENCODING_TRANSFORMS[transform_name]
             transform = klass()
-            restored_fit = {}
-            for serialization_key in klass.serialization_keys:
-                csv_path = join(
-                    models_dir,
-                    "%s.%s.csv" % (transform_name, serialization_key))
-                if exists(csv_path):
-                    restored_fit[serialization_key] = pandas.read_csv(
-                        csv_path, index_col=0)
-            if restored_fit:
+            target_path = join(models_dir, "%s.npz" % transform_name)
+            if exists(target_path):
+                with numpy.load(target_path) as loaded:
+                    restored_fit = dict(
+                        (key, loaded[key]) for key in loaded.keys())
                 if set(restored_fit) != set(klass.serialization_keys):
                     logging.warning(
                         "Missing some allele encoding transform serialization "
@@ -1116,12 +1109,11 @@ class Class1AffinityPredictor(object):
         ----------
         list of array
         """
-        loaded = numpy.load(filename)
-        weights = [
-            loaded["array_%d" % i]
-            for i in range(len(loaded.keys()))
-        ]
-        loaded.close()
+        with numpy.load(filename) as loaded:
+            weights = [
+                loaded["array_%d" % i]
+                for i in range(len(loaded.keys()))
+            ]
         return weights
 
     def calibrate_percentile_ranks(