PyPi Release

build/lib/epispot/__init__.py

@@ -8,13 +8,15 @@
 # imports
 import warnings
 import random
+import numpy as np
 from matplotlib import pyplot as plt
 from matplotlib import colors
 
 # local
 from . import comps
 from . import models
 from . import plots
+from . import fitters
 
 # version info
-__version__ = 'v1.0.0'
+__version__ = 'v1.1.0'

build/lib/epispot/fitters.py

@@ -0,0 +1,82 @@
+"""
+This module contains all available fitting algorithms.
+These operate separately from the `Model` class.
+STRUCTURE:
+    - grad_des
+"""
+
+from . import np
+
+
+def grad_des(get_model_pred, real_data, model_params, mu, epochs, N, samples, delta=0.0001):
+    """
+    The gradient descent fitter. This is not stochastic.
+    For long timespans, this may take a long time to converge.
+
+    :param get_model_pred: function in the form:
+                                param: parameters, parameters for model
+                                return: model predictions (use the integrate() method)
+    :param real_data: .csv file in the form:
+                        HEADER: ...layer #s
+                        CONTENTS: ...people in layers
+    :param model_params: list of model parameters
+    :param mu: the learning rate (adjust so that loss decreases after every epoch)
+    :param epochs: number of training sessions to run
+                   (more epochs = more accuracy + more time running)
+    :param N: the total population
+    :param samples: array of timestamps to use for training
+    :param delta: =0.0001, use small values for more precision--increase if gradients are 0
+    :return: optimized `model_params`
+    """
+
+    # get real data
+    file = real_data.readlines()
+    layers_to_opt = [int(char) for char in file[0].split(',')]
+
+    # quadratic cost
+
+    def cost(pred, real):
+
+        p_samp = []
+        r_samp = []
+
+        for sample in samples:
+            p_samp.append(pred[sample][1] / N)
+            r_samp.append(real[sample][1] / N)
+
+        return np.sum((np.array(p_samp) - np.array(r_samp)) ** 2)
+
+    data = [file[l].split(',') for l in range(1, len(file))]
+    for line in range(len(data)):
+        for char in range(len(data[line])):
+            data[line][char] = float(data[line][char])
+    for char in range(len(data[0])):
+        data[0][char] = int(data[0][char])
+
+    # training iterations
+    for epoch in range(epochs):
+
+        predictions = get_model_pred(model_params)
+        base_cost = cost(predictions, data[1:])
+        print('Epoch %s: Loss at %s' % (epoch, base_cost))
+
+        gradients = []
+
+        # parameter gradients
+        for param in range(len(model_params)):
+
+            model_params[param] += delta
+            pred = get_model_pred(model_params)
+
+            gradients.append((cost(pred, data[1:]) - base_cost) / delta)
+
+            model_params[param] -= delta
+
+        # print(gradients)
+        # print(model_params)
+        model_params = model_params - mu * np.array(gradients)
+        # print(model_params)
+        # predictions = get_model_pred(model_params)
+        # print(cost(predictions, data[1:]))
+
+    return model_params

epispot.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: epispot
-Version: 1.0.0
+Version: 1.1.0
 Summary: A tool for modelling infectious diseases.
 Home-page: https://quantum9innovation.github.io/epispot
 Author: quantum9innovation

epispot.egg-info/SOURCES.txt

@@ -2,6 +2,7 @@ README.md
 setup.py
 epispot/__init__.py
 epispot/comps.py
+epispot/fitters.py
 epispot/models.py
 epispot/plots.py
 epispot.egg-info/PKG-INFO

epispot.egg-info/requires.txt

@@ -1 +1,2 @@
 matplotlib
+numpy

epispot/__init__.py

@@ -19,4 +19,4 @@
 from . import fitters
 
 # version info
-__version__ = 'v1.0.0'
+__version__ = 'v1.1.0'

setup.py

@@ -5,7 +5,7 @@
 
 setuptools.setup(
     name="epispot",
-    version="v1.0.0",
+    version="v1.1.0",
     author="quantum9innovation",
     description="A tool for modelling infectious diseases.",
     long_description=long_description,