updated tiger.py

tiger.py

@@ -24,7 +24,7 @@ GUIDE_LEN = 23
 CONTEXT_5P = 3
 CONTEXT_3P = 0
 TARGET_LEN = CONTEXT_5P + GUIDE_LEN + CONTEXT_3P
-UNIT_INTERVAL_MAP = 'exp-lin-exp'
+UNIT_INTERVAL_MAP = 'sigmoid'
 
 # reference transcript files
 REFERENCE_TRANSCRIPTS = ('gencode.v19.pc_transcripts.fa.gz', 'gencode.v19.lncRNA_transcripts.fa.gz')
@@ -130,12 +130,12 @@ def calibrate_predictions(predictions: np.array, num_mismatches: np.array, param
     return correction * predictions
 
 
-def transform_predictions(predictions: np.array, params: dict = None):
+def score_predictions(predictions: np.array, params: pd.DataFrame = None):
     if params is None:
-        with open('transform_params.pkl', 'rb') as f:
-            params = pickle.load(f)
+        params = pd.read_pickle('scoring_params.pkl')
 
     if UNIT_INTERVAL_MAP == 'sigmoid':
+        params = params.iloc[0]
         return 1 - 1 / (1 + np.exp(params['a'] * predictions + params['b']))
 
     elif UNIT_INTERVAL_MAP == 'min-max':
@@ -180,7 +180,7 @@ def get_on_target_predictions(transcripts: pd.DataFrame, model: tf.keras.Model,
         # get predictions
         lfc_estimate = model.predict(model_inputs, batch_size=BATCH_SIZE_COMPUTE, verbose=False)[:, 0]
         lfc_estimate = calibrate_predictions(lfc_estimate, num_mismatches=np.zeros_like(lfc_estimate))
-        scores = transform_predictions(lfc_estimate)
+        scores = score_predictions(lfc_estimate)
         predictions = pd.concat([predictions, pd.DataFrame({
             ID_COL: [index] * len(scores),
             TARGET_COL: target_seq,
@@ -310,7 +310,7 @@ def predict_off_target(off_targets: pd.DataFrame, model: tf.keras.Model):
         ], axis=-1)
     lfc_estimate = model.predict(model_inputs, batch_size=BATCH_SIZE_COMPUTE, verbose=False)[:, 0]
     lfc_estimate = calibrate_predictions(lfc_estimate, off_targets['Number of Mismatches'].to_numpy())
-    off_targets[SCORE_COL] = transform_predictions(lfc_estimate)
+    off_targets[SCORE_COL] = score_predictions(lfc_estimate)
 
     return off_targets.reset_index(drop=True)