Update intro and machine learning talks and PDFs

Author: Joshua Taillon

talks/01_python_hs_intro/01_python_hs_intro.ipynb

@@ -62,58 +62,6 @@
     "best available for the purpose.*"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "metadata": {
-    "slideshow": {
-     "slide_type": "skip"
-    }
-   },
-   "source": [
-    "#### Import hyperspy"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 22,
-   "metadata": {
-    "slideshow": {
-     "slide_type": "skip"
-    }
-   },
-   "outputs": [],
-   "source": [
-    "%matplotlib notebook\n",
-    "import hyperspy.api as hs\n",
-    "import numpy as np"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "metadata": {
-    "slideshow": {
-     "slide_type": "skip"
-    }
-   },
-   "source": [
-    "#### Disable warnings for presentation:"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 2,
-   "metadata": {
-    "slideshow": {
-     "slide_type": "skip"
-    }
-   },
-   "outputs": [],
-   "source": [
-    "import logging\n",
-    "hs_logger = logging.getLogger('hyperspy') \n",
-    "hs_logger.setLevel(logging.ERROR)"
-   ]
-  },
   {
    "cell_type": "markdown",
    "metadata": {
@@ -556,12 +504,57 @@
    "metadata": {},
    "source": [
     "- Many algorithms exist to solve:   $\\mathbf{D}_{\\left(x,y\\right),E}=\\mathbf{W}_{\\left(x,y\\right)} \\times \\mathbf{S}_E$\n",
-    "  - Assumptions implicit in each affect their suitability for EDS, EELS, etc.\n",
-    "\n",
+    "  - This is a highly underdetermined problem, with an infinite number of solutions\n",
+    "  - Assumptions implicit in each algorithm affect their suitability for EDS, EELS, etc."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "slideshow": {
+     "slide_type": "fragment"
+    }
+   },
+   "source": [
     "- Primary methods (built into &nbsp;<img src='img/hyperspy.svg' width=40px style='display:inline-block;'/> HyperSpy ):\n",
-    "  - Principal component analysis (PCA) &mdash; finds non-physical spectra that describe the most variance in the datacube\n",
-    "  - Independent component analysis (ICA) &mdash; maximizes independence between spectral results\n",
-    "  - Non-negative matrix factorization (NMF) &mdash; enforces positivity in spectral components and weights"
+    "  - Principal component analysis ([PCA](https://en.wikipedia.org/wiki/Principal_component_analysis)) &mdash; finds non-physical spectra that describe the most variance in the datacube\n",
+    "  - Independent component analysis ([ICA](https://en.wikipedia.org/wiki/Independent_component_analysis)) &mdash; maximizes independence between spectral results\n",
+    "  - Non-negative matrix factorization ([NMF](https://en.wikipedia.org/wiki/Non-negative_matrix_factorization)) &mdash; enforces positivity in spectral components and weights"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "slideshow": {
+     "slide_type": "subslide"
+    }
+   },
+   "source": [
+    "## Principle component analysis + blind source separation"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "- PCA will find the spectra that describe the maximum amount of variance within the data set\n",
+    "    - These principle components (PCs) will contain the most \"information\" &mdash; but not information relevant to a microscopist\n",
+    "    - The results will look like vaguely spectral, but almost guaranteed to be non-physical"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "slideshow": {
+     "slide_type": "fragment"
+    }
+   },
+   "source": [
+    "- [Blind source separation](https://en.wikipedia.org/wiki/Signal_separation) often gives more interpretable results:\n",
+    "  - Independent component analysis (ICA):\n",
+    "      - Operates on the PCs to maximize independence between spectral results\n",
+    "  - Non-negative matrix factorization (NMF):\n",
+    "      - Decompose the matrix with constraints that match physical expectations"
    ]
   },
   {
@@ -596,8 +589,8 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "- Data collected from 0 to 360 degrees tilt in increments of 5 degrees\n",
-    "  - Dataset is 165 x 124 x 73 x 900    "
+    "- Data collected from 0 to 360 degrees tilt in increments of 5 degrees (73 spectrum images)\n",
+    "  - Dataset is $\\left(x, y, \\theta, E\\right)$ : $\\left(165, 124, 73, 900\\right)$ &mdash; 12.2GB"
    ]
   },
   {
@@ -676,9 +669,9 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "<div class='two-cols pro-con' style='font-size:1.5em;top:0'>\n",
+    "<div class='two-cols pro-con' style='font-size:1.3em;top:0'>\n",
     "\n",
-    "<div class='col' style='max-width:45%;margin-top:8%;'>\n",
+    "<div class='col' style='max-width:45%;margin-top:6%;'>\n",
     "<ul style='line-height:150%;'>\n",
     "    <li>PCA orders components by \"described variance\"</li>\n",
     "    <li><em>a priori</em> we know there should be three components</li>\n",
@@ -728,11 +721,32 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "<center><img style=\"width:65%;\" class=\"plain\" src=\"img/andy_eds_intensities_raw_vs_PCA.svg\"/></center>\n",
+    "<center><img style=\"width:65%;margin-top:5%;\" class=\"plain\" src=\"img/andy_eds_intensities_raw_vs_PCA.svg\"/></center>\n",
     "\n",
     "<div class=\"fig-caption-center\" style=\"height:4em; bottom:-3%;\">\n",
     "Line intensities extracted from model with top 3 PCA elements;<br/>\n",
-    "Drastically enhances S/N ratio in \"loading\" maps</div>"
+    "Drastically enhances S/N ratio in line intensity maps</div>"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {
+    "slideshow": {
+     "slide_type": "subslide"
+    }
+   },
+   "source": [
+    "## Result of PCA + BSS"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "<center><img style=\"width:70%;\" class=\"plain\" src=\"img/andy_BSS.svg\"/></center>\n",
+    "<div class=\"fig-caption-center\" style=\"height:4em; bottom:-3%;\">\n",
+    "One component for each element (phase)<br/>\n",
+    "Much more interpretable results</div>"
    ]
   },
   {
@@ -768,7 +782,9 @@
     "\n",
     "# <center>Thank you!</center>\n",
     "\n",
-    "<div style='text-align:center;margin:auto;'>Onto the demo...</div>"
+    "<div style='text-align:center;margin:auto;'>Onto the demo...<br/>Any questions?</div>\n",
+    "\n",
+    "<div style='text-align:center;margin:auto;margin-top:15px;'><a href=\"mailto:[email protected]\">[email protected]</a></div>"
    ]
   }
  ],