keras activations in different place give different results

rnn-activations.ipynb

{
  "cells": [
    {
      "metadata": {
        "ExecuteTime": {
          "end_time": "2018-09-02T05:41:35.125566Z",
          "start_time": "2018-09-02T05:41:35.118945Z"
        },
        "trusted": true
      },
      "cell_type": "code",
      "source": "inpdim,outdim,batch,timesteps = 2, 3, 8, 4\n\nx = np.random.rand(batch, timesteps, inpdim)\n\ny = np.random.randint(0, outdim, size=batch*timesteps)\ny = y.reshape(batch, timesteps, 1)",
      "execution_count": 8,
      "outputs": []
    },
    {
      "metadata": {
        "ExecuteTime": {
          "end_time": "2018-09-02T05:41:35.809413Z",
          "start_time": "2018-09-02T05:41:35.655003Z"
        },
        "trusted": true
      },
      "cell_type": "code",
      "source": "model1 = Sequential([\n    SimpleRNN(outdim, input_shape=(timesteps,inpdim), return_sequences=True, activation='softmax')\n])\n\nmodel1.compile(loss='categorical_crossentropy',optimizer=sgd())",
      "execution_count": 9,
      "outputs": []
    },
    {
      "metadata": {
        "ExecuteTime": {
          "end_time": "2018-09-02T05:41:36.982000Z",
          "start_time": "2018-09-02T05:41:36.694834Z"
        },
        "cell_style": "center",
        "trusted": true
      },
      "cell_type": "code",
      "source": "model2 = Sequential([\n    SimpleRNN(outdim, input_shape=(timesteps,inpdim), return_sequences=True),\n    Activation('softmax')\n])\n\nmodel2.compile(loss='categorical_crossentropy',optimizer=sgd())\nmodel2.set_weights(model1.get_weights())",
      "execution_count": 10,
      "outputs": []
    },
    {
      "metadata": {
        "ExecuteTime": {
          "end_time": "2018-09-02T05:41:37.705218Z",
          "start_time": "2018-09-02T05:41:37.493801Z"
        },
        "trusted": true
      },
      "cell_type": "code",
      "source": "assert (model1.predict(x) == model2.predict(x)).all()",
      "execution_count": 11,
      "outputs": [
        {
          "ename": "AssertionError",
          "evalue": "",
          "output_type": "error",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mAssertionError\u001b[0m                            Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-11-e4885fc5d345>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0;32massert\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mmodel1\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0mmodel2\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mall\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;31mAssertionError\u001b[0m: "
          ]
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "",
      "execution_count": null,
      "outputs": []
    }
  ],
  "metadata": {
    "gist": {
      "id": "6142c21a1466eeec98e85eb4ba5dd3c9",
      "data": {
        "description": "keras activations in different place give different results",
        "public": true
      }
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3",
      "language": "python"
    },
    "language_info": {
      "name": "python",
      "version": "3.6.6",
      "mimetype": "text/x-python",
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "pygments_lexer": "ipython3",
      "nbconvert_exporter": "python",
      "file_extension": ".py"
    },
    "_draft": {
      "nbviewer_url": "https://gist.github.com/6142c21a1466eeec98e85eb4ba5dd3c9"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 2
}