Merge pull request #51 from theGreatHerrLebert/david@mixture

adding predictor structure to algorithm

imspy/imspy/algorithm/rt/predictors.py

@@ -0,0 +1,29 @@
+import numpy as np
+import tensorflow as tf
+import importlib.resources as resources
+
+"""
+def get_model_path(model_name: str) -> str:
+    return resources.files('ionmob.pretrained_models').joinpath(model_name)
+
+
+def get_gru_predictor(model_name: str = 'GRUPredictor') -> tf.keras.models.Model:
+    return tf.keras.models.load_model(get_model_path(model_name))
+"""
+
+
+class ProjectToInitialSqrtCCS(tf.keras.layers.Layer):
+    """
+    Simple sqrt regression layer, calculates ccs value as linear mapping from mz, charge -> ccs
+    """
+
+    def __init__(self, slopes, intercepts):
+        super(ProjectToInitialSqrtCCS, self).__init__()
+        self.slopes = tf.constant([slopes])
+        self.intercepts = tf.constant([intercepts])
+
+    def call(self, inputs, **kwargs):
+        mz, charge = inputs[0], inputs[1]
+        # since charge is one-hot encoded, can use it to gate linear prediction by charge state
+        return tf.expand_dims(tf.reduce_sum((self.slopes * tf.sqrt(mz) + self.intercepts) * tf.squeeze(charge), axis=1),
+                              1)