@@ -2,5 +2,5 @@ include settings.ini
 include LICENSE
 include CONTRIBUTING.md
 include README.md
-include tklfp/*.pkl
+include tklfp/*.csv
 recursive-exclude * __pycache__
@@ -46,7 +46,7 @@ The package uses [parameters from the original 2020 paper](https://github.com/si
 tklfp = TKLFP(..., params=new_params)
 ```
 
-The new params must have the same content as the default [`tklfp.params2020`](https://github.com/siplab-gt/tklfp/blob/master/tklfp/__init__.py#:~:text=_sig_i%20%3D%202.1-,params2020%20%3D,-%7B). The `A0_by_depth` params are scipy interpolation objects, but could theoretically be any callable that will return A0 (in μV) for an arbitrary depth (in mm).
+The new params must have the same content as the default [`tklfp.params2020`](https://github.com/siplab-gt/tklfp/blob/master/tklfp/__init__.py#:~:text=_sig_i%20%3D%202.1-,params2020%20%3D,-%7B). The `exc_A0_by_depth` and `inh_A0_by_depth` params are scipy interpolation objects built dynamically from the package's CSV data, but could theoretically be any callable that will return A0 (in μV) for an arbitrary depth (in mm).
 
 ## Development
 For development, you can install the package in editable mode with development dependencies:
 
@@ -48,32 +48,19 @@
     }
    ],
    "source": [
+    "from tklfp import read_data, clean_data\n",
     "\n",
-    "def read_data(fname):\n",
-    "    with open(fname) as csvfile:\n",
-    "        r = csv.reader(csvfile)\n",
-    "        rows = []\n",
-    "        for row in r:\n",
-    "            rows.append(row)\n",
-    "        return np.array(rows, dtype=float)\n",
-    "\n",
-    "def clean_data(depth_amp, table_amp_values):\n",
-    "    depth_amp = np.vstack((depth_amp, [800, 0]))\n",
-    "    depth_amp[:, 0] = np.around(depth_amp[:, 0], -1)\n",
-    "    depth_amp[[0, 4, 8, 12], 1] = table_amp_values\n",
-    "    return depth_amp\n",
-    "\n",
-    "exc_depth_amp = read_data('../web-plot-digitizer/fig5a-points.csv')\n",
+    "exc_depth_amp = read_data(\"../tklfp/fig5a-points.csv\")\n",
     "exc_depth_amp = clean_data(exc_depth_amp, [-1.6, 4.8, 2.4, -0.8])\n",
     "\n",
-    "inh_depth_amp = read_data('../web-plot-digitizer/fig5b-points.csv')\n",
+    "inh_depth_amp = read_data(\"../tklfp/fig5b-points.csv\")\n",
     "inh_depth_amp = clean_data(inh_depth_amp, [-2, 30, -12, 3])\n",
     "\n",
     "fig, axs = plt.subplots(1, 2, sharey=True, figsize=(6, 2.5))\n",
     "axs[0].scatter(exc_depth_amp[:, 0], exc_depth_amp[:, 1])\n",
-    "axs[0].set(title='Excitatory uLFP', ylabel='uLFP amplitude (μV)', xlabel='Depth (μm)')\n",
+    "axs[0].set(title=\"Excitatory uLFP\", ylabel=\"uLFP amplitude (μV)\", xlabel=\"Depth (μm)\")\n",
     "axs[1].scatter(inh_depth_amp[:, 0], inh_depth_amp[:, 1])\n",
-    "axs[1].set(title='Inhibitory uLFP', xlabel='Depth (μm)')\n",
+    "axs[1].set(title=\"Inhibitory uLFP\", xlabel=\"Depth (μm)\")\n",
     "plt.show()"
    ]
   },
@@ -83,11 +70,11 @@
    "metadata": {},
    "source": [
     "## Some modifications\n",
-    "The amplitude values in the figure are 10 times what is recommended in table 1, so we will scale them down.\n",
-    "\n",
-    "To make the profile more flexible, we'll also interpolate and extrapolate, assuming that by 200 μm outside the reported range the uLFP drops to 0.\n",
-    "\n",
-    "Finally, we'll change from μm to mm for consistency with horizontal coordinates."
+    "The package preprocesses the data a little more and interpolates.\n",
+    "Preprocessing steps:\n",
+    "- The amplitude values in the figure are 10 times what is recommended in table 1, so we scale them down.\n",
+    "- To make the profile more flexible, we'll also interpolate and extrapolate, assuming that by 200 μm outside the reported range the uLFP drops to 0.\n",
+    "- Finally, we change from μm to mm for consistency with horizontal coordinates."
    ]
   },
   {
@@ -108,31 +95,16 @@
     }
    ],
    "source": [
-    "from scipy.interpolate import interp1d\n",
-    "def modify_and_interpolate_data(data):\n",
-    "    data_mod = data.copy()\n",
-    "    data_mod[:, 1] /= 10\n",
-    "    npad = 10  # adding multiple zeros will prevent spline interp overshoot\n",
-    "    deep_padding = np.zeros((npad, 2))\n",
-    "    deep_padding[:, 0] = np.linspace(-700, -600, npad)\n",
-    "    sup_padding = np.zeros((npad, 2))\n",
-    "    sup_padding[:, 0] = np.linspace(1000, 1100, npad)\n",
-    "    data_mod = np.vstack([deep_padding, data_mod, sup_padding])\n",
-    "    data_mod[:, 0] /= 1000\n",
-    "    return interp1d(data_mod[:, 0], data_mod[:, 1], kind='cubic', \n",
-    "                    fill_value=0, bounds_error=False)\n",
-    "\n",
-    "f_exc = modify_and_interpolate_data(exc_depth_amp)\n",
-    "f_inh = modify_and_interpolate_data(inh_depth_amp)\n",
+    "from tklfp import f_exc, f_inh\n",
     "\n",
     "fig, axs = plt.subplots(1, 2, sharey=True, figsize=(6, 2.5))\n",
     "depths = np.linspace(-1, 1.4)\n",
-    "axs[0].scatter(exc_depth_amp[:, 0]/1000, exc_depth_amp[:, 1]/10)\n",
-    "axs[0].set(title='Excitatory uLFP', ylabel='uLFP amplitude (μV)', xlabel='Depth (mm)')\n",
+    "axs[0].scatter(exc_depth_amp[:, 0] / 1000, exc_depth_amp[:, 1] / 10)\n",
+    "axs[0].set(title=\"Excitatory uLFP\", ylabel=\"uLFP amplitude (μV)\", xlabel=\"Depth (mm)\")\n",
     "axs[0].plot(depths, f_exc(depths), lw=1)\n",
-    "axs[1].scatter(inh_depth_amp[:, 0]/1000, inh_depth_amp[:, 1]/10, label='data')\n",
-    "axs[1].set(title='Inhibitory uLFP', xlabel='Depth (mm)')\n",
-    "axs[1].plot(depths, f_inh(depths), lw=1, label='interp')\n",
+    "axs[1].scatter(inh_depth_amp[:, 0] / 1000, inh_depth_amp[:, 1] / 10, label=\"data\")\n",
+    "axs[1].set(title=\"Inhibitory uLFP\", xlabel=\"Depth (mm)\")\n",
+    "axs[1].plot(depths, f_inh(depths), lw=1, label=\"interp\")\n",
     "axs[1].legend()\n",
     "plt.show()"
    ]
@@ -142,20 +114,7 @@
    "id": "414972a3",
    "metadata": {},
    "source": [
-    "This looks reasonable. Now let's save the interpolation objects:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "5e918b8d",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import pickle\n",
-    "with open('../tklfp/exc_A0_by_depth.pkl', 'wb') as exc_file, open('../tklfp/inh_A0_by_depth.pkl', 'wb') as inh_file:\n",
-    "    pickle.dump(f_exc, exc_file)\n",
-    "    pickle.dump(f_inh, inh_file)"
+    "This looks reasonable."
    ]
   }
  ],
 
@@ -1,21 +1,57 @@
 __version__ = "0.3.0"
 """Lightweight implementation of Telenczuk 2020 kernel LFP approximation"""
-import pickle
+import csv
 from typing import Union
-
 import numpy as np
-import scipy  # noqa: F401
 from importlib_resources import files
 from numpy.typing import ArrayLike
+from scipy.interpolate import interp1d
+
+
+def read_data(fname):
+    try:
+        with files(__package__).joinpath(fname).open("r") as csvfile:
+            r = csv.reader(csvfile)
+            rows = []
+            for row in r:
+                rows.append(row)
+            return np.array(rows, dtype=float)
+    except (FileNotFoundError, ModuleNotFoundError, ValueError):
+        with open(fname, "r") as csvfile:
+            r = csv.reader(csvfile)
+            rows = []
+            for row in r:
+                rows.append(row)
+            return np.array(rows, dtype=float)
+
+
+def clean_data(depth_amp, table_amp_values):
+    depth_amp = np.vstack((depth_amp, [800, 0]))
+    depth_amp[:, 0] = np.around(depth_amp[:, 0], -1)
+    depth_amp[[0, 4, 8, 12], 1] = table_amp_values
+    return depth_amp
+
 
+def _load_uLFP_A0_profile(fname, table_amp_values):
+    depth_amp = read_data(fname)
+    depth_amp = clean_data(depth_amp, table_amp_values)
 
-def _load_uLFP_A0_profile(fname):
-    # with open_binary(__package__, fname) as f:
-    with files(__package__).joinpath(fname).open("rb") as f:
-        A0_profile = pickle.load(f)
+    data_mod = depth_amp.copy()
+    data_mod[:, 1] /= 10
+    npad = 10  # adding multiple zeros will prevent spline interp overshoot
+    deep_padding = np.zeros((npad, 2))
+    deep_padding[:, 0] = np.linspace(-700, -600, npad)
+    sup_padding = np.zeros((npad, 2))
+    sup_padding[:, 0] = np.linspace(1000, 1100, npad)
+    data_mod = np.vstack([deep_padding, data_mod, sup_padding])
+    data_mod[:, 0] /= 1000
+    return interp1d(
+        data_mod[:, 0], data_mod[:, 1], kind="cubic", fill_value=0, bounds_error=False
+    )
 
-    return A0_profile
 
+f_exc = _load_uLFP_A0_profile("fig5a-points.csv", [-1.6, 4.8, 2.4, -0.8])
+f_inh = _load_uLFP_A0_profile("fig5b-points.csv", [-2, 30, -12, 3])
 
 ######### PARAMETERS #############
 _sig_i = 2.1
@@ -25,8 +61,8 @@ def _load_uLFP_A0_profile(fname):
     "sig_i_ms": _sig_i,  # std-dev of ihibition (in ms)
     "sig_e_ms": 1.5 * _sig_i,  # std-dev for excitation
     "d_ms": 10.4,  # constant delay
-    "exc_A0_by_depth": _load_uLFP_A0_profile("exc_A0_by_depth.pkl"),
-    "inh_A0_by_depth": _load_uLFP_A0_profile("inh_A0_by_depth.pkl"),
+    "exc_A0_by_depth": f_exc,
+    "inh_A0_by_depth": f_inh,
 }