linear model with NAL data kristoferB/SP#86

models/linear/.gitignore

@@ -0,0 +1,3 @@
+NALState2017-07-11T15:14:43.994+02:00.csv
+modeldir/
+exportedmodel/

models/linear/train_and_export_model.py

@@ -0,0 +1,40 @@
+import tensorflow as tf
+from tensorflow.contrib.learn import LinearRegressor
+from tensorflow.contrib import layers
+from tensorflow.contrib.learn.python.learn.utils import input_fn_utils
+tf.logging.set_verbosity(tf.logging.INFO)
+
+import pandas as pd
+
+FEATURES = ["ttt30", "all", "MEP", "triaged", "PRIO3", "PRIO4"]
+
+pdframe = pd.read_csv("NALState2017-07-11T15:14:43.994+02:00.csv")
+
+def input_fn_train():
+    feature_cols = {name: tf.constant(pdframe[name].get_values()[:-1]) for name in FEATURES}
+    # TODO atm predicting next ttt30 when sample intervals are 23 min..
+    outputs = tf.constant(pdframe["ttt30"].get_values()[1:])
+    return feature_cols, outputs
+
+feature_cols = [layers.real_valued_column(name) for name in FEATURES]
+regressor = LinearRegressor(
+    feature_columns=feature_cols,
+    model_dir="./modeldir"
+)
+regressor.fit(input_fn=input_fn_train, steps=10000)
+
+'''
+def serving_input_fn():
+    default_inputs = {col.name: tf.placeholder(col.dtype, [None]) for col in feature_cols}
+    features = {key: tf.expand_dims(tensor, -1) for key, tensor in default_inputs.items()}
+    return input_fn_utils.InputFnOps(
+        features=features,
+        labels=None,
+        default_inputs=default_inputs
+    )
+
+regressor.export_savedmodel(
+   "exportedmodel",
+    serving_input_fn
+)
+'''